Citation
An Acoustical/temporal analysis of emotional stress in speech

Material Information

Title:
An Acoustical/temporal analysis of emotional stress in speech
Creator:
Hicks, James Woodrow, 1950- ( Dissertant )
Hollien, Harry ( Thesis advisor )
Rothman, Howard B. ( Reviewer )
Brown, William S. ( Reviewer )
Paige, Arnold ( Reviewer )
Gerhardt, Kenneth ( Reviewer )
Place of Publication:
Gainesville, Fla.
Publisher:
University of Florida
Publication Date:
Copyright Date:
1979
Language:
English
Physical Description:
xx, 160 leaves : graphs ; 28 cm.

Subjects

Subjects / Keywords:
Combined stress ( jstor )
Critical loading ( jstor )
Emotional states ( jstor )
Lexical stress ( jstor )
Psychological stress ( jstor )
Standard deviation ( jstor )
Stress distribution ( jstor )
Stress ratio ( jstor )
Stress tests ( jstor )
Voice stress analyzers ( jstor )
Dissertations, Academic -- Speech -- UF ( lcsh )
Speech ( lcsh )
Speech thesis Ph. D ( lcsh )
Stress (Psychology) ( lcsh )
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

Notes

Abstract:
The external manifestation of emotions has been of interest to researchers for many years. For example, as early as 1872 Charles Darwin discussed the use of facial and body movements as indicators of emotion in his book The expression of the Emotions in Man and Animals . However, Darwin's text is only descriptive in nature. Subsequently, various physiological measures have been correlated to emotional states; among these measures have been (1) heart rate, (2) galvanic skin response (GSR), (3) electroencephalogram (EEG) , (4) respiration and (5) blood pressure. However, more recently scientists have investigated speech parameters as possible indicators of the emotional state of the speaker. There are many instances in which knowledge about the emotional state of an individual would be desirable. It also would be advantageous to obtain this information without any direct, physical contact with the individual—as is currently required with the above mentioned physiological measures. A research thrust aimed at determining the acoustical/temporal speech parameters correlated to emotional stress should be useful to the area of speech communications. Most of the previous investigations have been limited in scope. For instance, many have analyzed only one parameter—primarily fundamental frequency (f ) —and/or had very small subject populations . The thrust of the current study focuses on an acoustical/ temporal analysis of the effects of stress on speech. Included in the study were both laboratory induced stress and a "real" situational stress. The laboratory stress was induced via an electrical shock as the stressor. The situational stress- environment consisted of a university level public speaking course in which the students were recorded while delivering speeches to an audience of their peers Included in the analysis were several characteristics within the intensity, fundamental frequency and temporal domains. The following parameters were analyzed within the intensity domain: (1) maximum, (2) mean and (3) mode of the intensity distribution. The mean and distribution of speaking fundamental frequency (SFF) were analyzed within the frequency domain. The temporal parameters included: (1) speech/pause ratio, (2) speech rate, (3) the time energy distribution, (4) the number of speech bursts, (5) the number of pauses, and (6) speech time/total time ratio. In addition, the number of disfluencies in each speech sample also was measured. A comparison was made between speech samples produced normally and under stress. The comparison was based on the acoustical/temporal parameters analyzed. The results of the data analysis indicate that measurable acoustical/temporal changes do occur in speech produced under stress as compared to normal, non-street speech. However, the magnitude of these changes seems to be a function of the type of stress. For the situational stress experiment, 68.8% of the parameters differed significantly between the normal and stress speaking conditions, whereas only 18.8% changed significantly in the laboratory stress experiment . It was found that for the laboratory stress paradigm all of the parameters increased for the stress speech, relative to normal. However, only the maximum intensity, time-energy distribution and the number of disfluencies increased significantly. That is, the stress induced by electro-shock did not alter the subjects' speech patterns to any great extent. However, significant differences were found to exist between the normal and stress speech samples in the situational stress experiment. Specifically, significant decreases were found for the intensity measures as well as the number of speech bursts and pauses. Conversely, the mean SFF and temporal parameters (except for speech rate) were found to increase significantly. Based on these findings, the general effects of stress on speech seem to: (1) decrease intensity, (2) increase SFF, (3) slightly decrease snpf>ch rat-.e. and (4) decrease? the number of speech bursts and pauses, resulting in longer speech bursts. However, the variability in these parameters also increased indicating that the observed changes may not be uniformly consistent for all individuals. Therefore, baseline (normal) data maybe required for. an individual before the presence of stress in that individual can be detected.
Thesis:
Thesis--University of Florida.
Bibliography:
Bibliography: leaves 154-158.
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
by James Woodrow Hicks, Jr.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
023343007 ( AlephBibNum )
06572942 ( OCLC )
AAL2929 ( NOTIS )

Downloads

This item has the following downloads:


Full Text










AN ACOUSTICAL/TEMPORAL ANALYSIS OF
EMOTIONAL STRESS IN SPEECH










By

JAM1ES WOODROW HICKS, JR.


A DISSERTATION PRESENTED TO THIE GRADUATE COUNCIL OF TIHE
UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THIE DEGREE OF
DOCTOR OF PH~LOSOPHY


UNIVERSITY OF FLORIDA
1979





























Copyright 1979

by

James Woodrow Hicks, Jr.































This dissertation is dedicated

to the memory of my parents

James and Evelina Hicks














ACKNOWLEDGMENTS


I would like to express special thanks to Dr. Harry

Hollien for his encouragement, guidance and help through

some very trying times. He also counseled me during per-

sonal problems, which is deeply appreciated and "motivated"

me when it was needed. Thanks are also extended to Drs.

How~ard Rothman, Arnold Paige, and W. Samuel Brown for taking

the time to be members of my committee.

Thanks are also due Dr. Williamn Cutler of the Gainesville

Veteran's Administration Hospital for the psychogalvanometer

used to administer the electroshock in the laboratory experi-

ment. Also, Dr. Tom Saine, Jim Harris, Ralph Partridge,

Dr. Don Williams and Don Stacks, of the Speech Department,

for allowing me to come into their classrooms to record

their ~public speaking classes, thank you.

I would like to acknowledge the entire staff of the

Institute for Advanced Study of the Communication Processes

(IASCP). None of us at IASCP would get much accomplished

without their help.









I would also like to thank all my friends for their

encouragement during my work on this dissertation. Thanks

also to all th-ose who volunteered as subjects, specially

for the electro-shock experiment, for my research.

There are too few words to express my thanks to Cindy

Dowey. She w~as always by my side and encouraged me in the

worst of times. Our relationship did the most in helping me

complete this dissertation.
















TABLE OF CONTENTS






ACKNOWIEDGMENTS .................. iv


LIST OF TABLES. .. .. .. .. .. .. .. .. x


LIST OF FIGURES .. .. ... .. .. .. .. xv


ABSTRACT. .. .. .. .. .. ... ... . xvil


CHAPTER


I INTRODUCTION. . .. .. .. ... . 1


psychology/Psychiatry. . .. .. .. 2
Forensic Communication ... .. 3
Diver Communication. .. .. .. .. 6
Aro~~space: Communiiications . .. .. .
A Definition of Stress .. .. .. 10


Laboratory Induced Stress . ... 12
Situational Stress. . .. .. .. 13


Measures of Stress . .. ... .. 14


Physiological Measures. .. .. 14
Acoustical/Temporal Measures
of Stress. .. .. .. ... 17
Intensity . .. .... .. 17
Fundamental Frequency (fo). 9
Temporal Measures . .. ... .. 23
Disfluency. .. .. .. .. .. 24
Subjective Judgements of Stress . 25

Aural perception .. .. .. 25
Self-rating by subjects. .. 26










TABLE OF CONTENTS (Continued)



Page

Objectives .. .. ... .. .. 27

II METHIOD. .. ... ... .. .. .. 30

Acoustical Analyses. ... .. .. 31


Speech Intenai~ty (Sl) ... .. 31
Speaking Fundamental Frequency
(SFF). .. . .. .. .. .. 34

Temporal Analysen ,. .. .. .. . 36

Speech Rate (n;R). ... . .. .. 36
Time-Energy Distribution (TED). . 37
Speech/Pause Ratio (SPR). .. .. 39
Speech Bursts and Pauses. .. .. 39
Speech Time/Tcital Time
(ST/TT) Ratio. . .. .. .. .. 40

Disfluency ... ~ . ... .. .. 41
Subjects' Self-R.;h'ing Stress .- . .. 42
Recording Procedure. .. .. . 43
Experiment I: Taboratory Stress .. 46

Population. .... .. . ..-. 47
Speech Materi.;l~s. .- .. . .. 47
Experimental P'rocedure. .. ... 48

Experiment II: fDituational Stress .. 49

Population. . .. .. .. .. 50
Speech Materiail~s ... .. .. .51
Experimental iLracedlure.. . .. 51

Statistical Analy:;is . .. ... 52

III RESULTS OF LABORATC~".v :;TRESS EXPERIMENT 55


Speech Intensity (SI). . ... .. 56
Speaking Fundamental Frequency (SFF) 63











TABLE OF CONTENTS (Continued)


Page

Speech Rate (SR) .. .. .. ... 65
Speech/Pause Ratio (SPR) --. .. . 70
Time-Energy Distribution (TED) . .. 72
Speech Time/Total Time Ratio (ST/TT) . 76
Speech Bursts and Pauses .. .. .. 78
Disfluency . . .. . . 78

IV RESULTS OF SITUATIONAL STRESS EXPERIMENT 82


Speech Intensity (SI). ... .. .. 83
Speaking Fundamental Frequency (SFF) . 90
Speech Rate (SIR) . .. .. . .. 92
Speech/Pause Ratio (SPR) . .. . .. 97
Time-Energy Distribution (TED) . . 98
Speech Time/Total Time Ratio (ST/TT) . 104
Speech Bursts and Pauses . .. .. 106
Disfluency .. .. .. .. .. . 106

V DISCUSSION .... . .. .. .. .. 110


Laboratory Stress Experiment . . .. 110
Situational Stress Experiment. . . 114
Comparison to Previous Studies: SPF . 118
Comparison to Previous Studies:
Intensity. . ... .. . .. 120
Comparison to Previous Studies:
Temporal. . . .. .. .. ... 121
Conclusions. .. .. .. . . 123

APPENDICES
APPENDIX A--MULTIPLE AFFECT ADJECTIVE
CHECK LIST .. .. ... . .. 128
APPENDIX B--SUBJECT INFORMED CONSENTr FOT~I~ , 131
APPENDIX C--INSTRUCTIONS TO SUBJECTS. .. .. 133
APPENDIX D--TABULATED VALUES FOR TIEF
INTENSITY DISTRIBUTIONS. .. .. 138
APPENDIX E--TABULATED VALUES FOR SFF
DISTRIBUTIONS . .. .. .. 145
APPENDIX F--TABULATED VALUES FOR TIM~E-ENEP VTr'
DISTRIBUTIONS ('IED). .. . 152


Vlll












TABLE OF CONTENTS (Continued)


Pacre


REFERENCES. .. .. . . .


BIOGRAPHICAL SKE~TCH














LIST OF TABLES


Table Pacy

1 Results of the Speech Intensity (SI)
Analyses for the Laboratory Stress Experl-
ment; nl~l Values are Expressed in dB Re:
Imy. Values in Parentheses are the Standard
Devia~iol (;SD) and Degress of Freedom (df)
for the t Scores. .. ... .. .. .. 57

2 Results; of the Speakcing Fundamental Fre-
quency (SFF) Analyses for the Laboratory
Stress Experiment; Values are Expressed
in Both Semitones (ST) and HIertz (Hz);
Values in Parentheses are the standard
Deviations (SD) and the Degrees of Freedom
(df) or:?? thle t Scores .. .. ... .. 64

3 Results of the Speech RaLe i5R) AnaiilySs
for Umr r~aboratory Stress Experiment; SR
Values are Expressed in Syllables Per
Second; Values in Parentheses are the
Standard Deviations (SD) and the Degrees
of Freedom~ (*df) for the t Scores
(N = 12). . .. .. .. .. . 69

4 Results of the Speech/Pause Ratio (SPR)
Analysis for the-Laboratory Stress Experi-
ment; Vralues in Parentheses are the Standard
Deviations (SD) and the Degrees of Freedom
(.df) fo~r the t Scores (N = 12). ... .. 71

5 Results of the Speech Time/Total Time (ST/TT)
Ratio Analilysis for the L~aboratory Stress
Experiment; Values in Parentheses'are the
Standard Docviations (SD) anld the Degree of
Freedom (df) for the t Scores (N = 12). .. 77








LIST OF TABLES (Continued)


Table Page

6 Results of the Speech Bursts alnd Pause
Analyses for the Laboratory Stress Ex-
periment; Value in Parentheses is the
Degrees of Freedom (-df) for t~he t
Scores (N = 12) - * -*-*. 79

7 Analysis Results of the Number of Dis-
fluencies for the Laboratory !;;ress
Experiment; Value in Parenthone~s is the
Degrees of Freedom (df) for thle t Score
(N=~12). ..-...--......--- 80

8 Results of the Speech Intensi;t~y (SI)
Analyses for the SituationaJ :;tress
Experiment; All Values are Xxpressed in
dB Re: Imy. Values in Parenitheses are the
Standard Deviations (SD) and- the Degrees
of Freedom (df) for the t Sc~,res. . ... 84

9) Results of the Speaking F'undanentcl-al
Frequency (SFF) Analyses for the Situa-

Expressed in Both Semitoner: (ST) and
Hertz (Hz); Values in Pareiithe~ses are the
Standard Deviations (SD) and~ Lhe Degrees
of Freedom (df) for the t Scores. .. .* 91

10 Results of the Speech Rate (SR) Analysis
for the Situational S'tress Experiment;
SR Values are Expressed in :Syllables Per
Second; Values in Parentheses are the
Standard Deviations (SD) andi the Degrees
of Freedom (.df) for the t Scores (N = 17) 96

11 Results of the speech/Pause Rabia (SPR)
Analysis for the Situational Stress Ex-
periment; Values in Parenthenes:c are the
Standard Deviations (SD) anid the Degrees
of Freedom (df) for the t :~nrrmo (N = 17) 99









LIST OF TABIEdS (Continued)


Table Pacy

12 ires-ults~ of the Speech Time/Total Timie
(ST/TT) Ratio Analysis for the Situa-
tional Stress Experiment; Values in-
Parenthieses are the Stundard Deviat~ions
(SDn) and th3 Degrees of Freedom (df)
for the t Scores (N = 17) . . ... 105

13 Resullts, of the Speech Bursts and rPauseF
Analyses for the Situational Stress
Experiment; Value in Parentheses i:;
the Degrees of Freedom (df) for the
L. Scores (N = 17) .. ... . .. . 107

14 Analysis Results of the Number of Dis-
f~luenrcies for the Situational Stress
Experiment; Value in Parentheses is the
Degrees of Freedom (df) for the t rScore
(N =: 1.7). .. ... ... 109

15 SummarydT of Results for the Laborariocy
.Ctress, E:xperiment p+ased on Overa:ll MeanO
V~'nes for Each Parameter .. .. .. .. Ill

16 ':lnri;aryy oflResults for the Situationll~
Stress Experiment Based on Overall Mean
Va!~lue for Each Parameter .. ... .. 115

17 Co~mpari~son of the Overall Results for
Lbe laboratory and Situational Stress
Experiments Based on Mean Values for
E~ach Parameter. ... .. .. - 119

D-1 Intensity Distribution Values for the
I'lale ;ub~jects in the laboratory :;lrress
E~xperiment; Intensity Values are Ex:-
pressed in dB re:1my. ... . .. .... 138

D-2 :11ha:r U.t~y ~is tribu~tion Values for- thei
I'eia.le Subjects in the Lalboratory GCress;
':'l''r inte~nt; Intensity Values arc I:
pressed in dB re:1my. ... .. .. .. 139










LIST OF~ TABLES (Continuedl)


Table Pajge

D-3 Intensity Distribution Values for All
Subjects in the Laboratory Stress Experi-
ment; Intensity values are Expressed in
dB re:1my . . . .... .. .. .. 140

D-4 Intensity Distribution Values for the
Male Subjects in the Siltuational Stress
Experiment: Intensity Vcllues are Expressed
in d13 re:1my. . ... .. .. .. .. 141


D-5 Intensity Distribution Values for the
Female Subjects in thei Situational Stress
Experiment; Intensity Values are Expressed
in dB re:1my. . .. . .. .. .. .. 142


D-6 Intensity Distribution Values for All
Subjects in the Situational Stress
Experiment; Intensity Values are Expressed
in dB re:1my. . .. .. .. . ... 143



Distribution for the Mi~le Subjects in
the Laboratory Stress Experiment; Values
of SFF are Expressed in Semitone (ST)
Intervals . ... .. . . . 145


E-2 Speaking Fundamental Fr'equency (SFF)
Distribution for the Female Subjects in
the Laboratory Stress Experiment; Values
of SFF are Expressed LII Semitone (ST)
Intervals . .. .. ... . .. 146


E-3 Speak~ing Fundamental Prequenrlcyc (SFF)
Distribution for All Subijects in the
Laboratory Stress ExperiTment; Values
of SFF are Expressed in Semitonoe (ST)
Intervals.. .. .. .. .. .. 147


X111










LIST OF TABLES (Continued~c)


Table Pag

E-4 Speaking Fundramlenta..L L Frequency (SPF)
Distribution for the Male Subjects in the
Situational :ilchros E~xperiment; values of
SFF are Expressed in Semitone (ST)
Intervals .. .. .... .. .. .. 148

E-5 Speaking Furltndamenta Frequencyy (SFF)
Distributioni for' the Female Subjects in the
Situational Sttress Experiment; Values of
SFF are Exputate:cd ini Semitone (ST)
Internals .. .. .. .~ . .. .. 19

E-6 Speaking Fund Distribution for: A.ll Subjects in the
Situational Sircus Experiment; Values
of SFF are Expresnsed in Semitone (ST)
Intervals . .. .. .. .. ... 150

F-1 Mean Values :'W: ;-hnr Time-Energy Distri-
butions (TED) fIor t-he La~boratory Stress

Time for Each ofr NI~nn Energy Levels . .. ]52

F-2 Mean Values 0:'e Piva~ Time-Energy Distri-
butions (TED) Irar th-e Situational Stress
Experiment; Values!c are, the Per Cenit of
Time for Each ofE N~ine Energy Levels .. 153














LIST OF FIGURES



F~igur rage

1 F;M W~ireless Microphone/Tranlsmitter Used
in Recording Process . .. . .. ... 44

2 -Specially Designed Headset Used to Hold FM
Microphlone/Transmitter (Shrown irn Place) in
a Fixed Position Relative to Speaker .. 45

3 Intensity Distribution for tL Male5
Subjects in the Laboratory St:ress
Experiment . ... . .. .. .. 60

4 Intensity Distribution for Lte Female
Subjects in the Laboratory Firess
Experiment .. .. . .. . ... 61

5 Ovsrall Intensityr Distribuition for All
Subjects in the Laboratory Ti-ress
Experiment .. .. .. . . .. .. 62

6 Speaking Fundamental Frequenc~y (bFF)
Distribution for the Male Sibljects in
itlhe Laboratory Stress Expe.irime~nt . .. 66

Speaking Fundamental Frequcen-y, (SFUL)
Distribution for the Female Subjects
jin the Laboratory Stress ExF riment.t. .. 67

8 Overall Speakcing Fundamental1 Prequency
(SFF) Distributionl Ear All n;ubjctsl: in
the Laboratory Stress Experiment . ... 68

Time-Energyg Distribution ('I'T:U) for~ the
Male~ Subjects in the Laboratory rStress
Experiment ... .. .. . ... .. 73

10 Time-Energy Distribution (TTED) f-or the
Female Subjects in the La~bor.atory Stress
Experiment .. .. .. .. . 74
.xv








LIST OF FIGURES (Continued)


v isur a


Page


11 Overall Time-Energy Distribution (T'ED)
for All Subjects in the Labortatory
Stress Experiment. .. .. . ... .. 75

12 Intensity Distribution for the Ma;le
Subjects in the Situational Stress
Experiment .. .. ... . .. .. 87

13 Intensity Distribution for the F'emale
Subjects in the Situational Strcn:;
Experiment .. .. .. . . .. .. 88

14 Orrerall Intensity Distribution inc Anll
Subjects in the Situational Str-ess; Ex-
periment ... .. . . .. .. 89

15 Speaking Fundamdntal Frequency (Dev'c)
Distribution for the Male Subjects in
the Situational Stress Experiment, . .. 93

16SPsaking Pindamental Freque~nry (SW)'
Distribution for the Femalo Subjectis
in the Situational Stress Experime~lnt .. 94

17 Overall Speaking Fundamental. F'requiency
(SFF) Distribution for All Subjects~ in
ilhc Situational Stress Experimenlt. .. .95)

18 TIimie-Energy Distribution (TED) .;.or "he
Male Subjects in the Situational Stress
Experiment .. .. .. .. .. .. .. 101

19 Time-Energy Distribution (TED) fn~r the
Mem~iale Subjects in thie Situantiion;
Stress Experiment. .. ... .. .. 102

20 Overa.ll TIime-E~nergy D~istribution TED
for All Subjects in the Situatio;Il
Stress Experiment. . ... ... .. 103


XVI








Abstract of Dissertation Presentcd to the Graduate Council
of thie University of Florida in P'artLial Fulfillment of the
Requirements for the Degree of Doctor of Ph~ilosophy

AN ACOUSTICAL/TEMPORAL ANALYSIS OF
EMOTIONAL STRESS IN SPEECH

By

James Woodrow Hicks, Jr.

December, 1979

Chairman: H-arry Hollion
Major Department: Speech Department

The external manifestation of emotions has been of in-

torest to researchers for many years. For example, as early

as 1872 Charles Darwin discussed the use of facial and body

movements as indicators of emotion in his book The.Expression

of the Emotions in Man? and Animals. However, Darwin's text

is only descriptive in nature. Subsequently, various

physiological measures have been correlated to emotional

states; among these measures have been (1) heart rate, (2)

galvanic skin response (GSR), (3) e lect~roencephnalogram

(EEG), (4) respiration and (5) blood pressure.

However, more recently scientists have investigated

speech pa~rametcor as possible indicators of the emotional

state of the speaker. There are many instances in which

knowledge about the emotional state of an individual would

be desirable. It also would be advantageous to obtain this


XVll









information without any direct, physical contact with the

individual--as is currently required with the above men-

tioned physiological measures. A research thrust aimed

atr determining the acoustical/temporal speech parameters

correlated to emotional stress should be useful to the

area of speech communications. Most of the previous in-

vestigations have been limited in scope. For instance,

many have analyzed only one parameter--primarily fundamen-

tal frequency (fo)--and/or had very small subject popula-

tions.

The thrust of the current study focuses on an acousti-

cal/temporal analysis of the effects of stress on speech.

Included in the study were both laboratory induced stress

and a "real" situational stress. The laboratory stress was

induced via an electrical shock as the stressor. The

situational stress environment: consisted of a university

level public speaking course in which the students were

recorded while delivering speeches to an audience of their

peers.

Included in the analysis were several characteristics

within the intensity, fundamental flrequency and temporal

domains. The following parameter:; were analyzed within the

inte ns ity doma in: (1) maximum, (2) mean and (3) mode of the

intensity distribution. The mean and distribution of

XVlll









speaking fundamental frequency (SFF) were analyzed within

the frequency domain. The temporal parameters included:

(1) speech/pause ratio, (2) speech rate, (3) the time-

energy distribution, (4) the number of speech bursts, (5)

the number of pauses, and (6) speech time/total time

ratio. In addition, the number of disfluencies in each

speech sample also was measured. A comparison was made

between speech samples produced normally and under stress.

The comparison was based on the acoustical/temporal parame-

ters analyzed.

The results of the data analysis indicate that measur-

able acoustical/temporal changes do occur in speech produced




However, the magnitude of these changes seems to be a func-

tionl of the type of stress. For the situational stress

experiment, 68.8%/ of the parameters differed significantly

between the normal and stress speaking conditions, whereas

only 18.8% changed significantly in the laboratory s-tress

experiment.

It was found that for the laboratory stress paradigm

all of the parameters increased for the stress speech,

relative to normal. However, only the maximum intensity,

time-energy distribution and the number of disfluencies









increased significantly. That is, the stress induced by

electro-shock did not alter the subjects' speech patterns

to any great extent.

However, significant differences were found to exist

between the normal and stress speech samples in the situa-

tional stress experiment. Specifically, significant de-

creases were found for the intensity measures as well as the

number of speech bursts and pauses. Conversely, the mean

SFF and temporal parameters (except for speech rate) were

found to increase significantly. Based on these findings,

the general effects of stress on speech seem to: (1) de-

crease intensity, (2) increase SFF, (3) slightly decrease

speech rate; and (4) decrease the number of speech bursts

and pauses, resulting in longer speech bursts. However, the

variability in these parameters also increased indicating that

the observed changes may not be uniformly consistent for all

individuals. Therefore, baseline (normal) data may be re-

quired for, an individual before the presence of, stress in,

that individual can be detected.














CHAPTER I


INTRODUCTION


For some time scientists have searched for parameters

within the speech signal that night reveal the emotional

state of the talker (Bonner, 1943; Cohen, 1961; Friedhoff

et al., 1964; Barland, 1973; Fairbanks & Hoaglin, 1941;

Fairbanks & Provnost,. 1939; Hecker et al., 1968; Kuroda

et al., 1976; Silverman & Silverman, 1975; Simonov & Frolov,

1973, 1977; Williams & Stevens, 1969, 1972). A number of




vances in technology; hence, some of the parameters these sci-

entistr may have preferred to study were not amenable to

their research. However, recent developments of equipment

and techniques for the acoustical analysis of speech permit

the investigation of issues previously not possible.

It is well known that speech contains both linguistic

and paralinguistic messages. That is, virtually any utter-

ance conveys information on several levels--including that

related to cultural background, regional dialect, country

of origin, and possibly the general feeling about the









statement being made by the speaker. Perhaps more impor-

tant, a talker can transmit information via speech that

reflects his current health and emotional state.

There are many instances in which knowledge about the

emotional state of an individual would be desirable. Further,

there are situations where it would be advantageous to obtain

this information without any direct, physical contact with

the person being evaluated--for example, in aerospace or

diving opera tions. Indeed, research directed at determining

the acoustical correlates of emotional stress should be use-

ful to many of the sub-specialties within the speech com-

munication area; among these specialties are Forensics,

I.-LUrcpce andl Diver Comm~iuniicaltionsl a5 Well as PsYchologU~Y/

Psychiatry.


Psychology/Psychiatry


Psychiatrists and psychologists appear to rely on the

parallel encoding of linguistic and emotional information by

their patients in order to treat these individuals (Friodhoff

et al., 196i4; Ostwald, 1963). For example, Ostwald states,

"whenever direct expression involves sound making--emotive

sound making--acoustics has something tangible to con-

tribute to psychotherapy" (p. 85). TIhus, it would. appear









that Ostwald recognized the contribution to be made to

psychotherapy by paralinguistic information encoded within

the speech signal. A substantial benefit would accrue to

psychotherapy if objective methods of evaluating the emo-

tional state of the patient could be developed and used

to complement subjective observations of the psychologist/

psychiatrist. In short, if the therapist could identify

the emotion (fear, anger, or anxiety, say) being expressed

by the patient, he should be able to provide more effective

treatment.


Forensic Communication





to the area of Forensic Communication. For example, when a

criminal is recorded making a bomb threat, it is desirable

to determine his emotional state: Is it one where there is

a high probability that he will follow through with the

threat? Another illustration: It is obvious that when a

suspect lies during interrogation, he is under stress.

Can the acoustic speech signal be used to determine the

likelihood that the suspect is lying in such cases? In

an effort to answer these questions, "detectors" of voice

stress recently have been developed. They are purported









to be effective as "lie detectors"; they include the

Psychological Stress Analyzer (PSA), Mark II, Hlagoth

(Bennett, 1977), Psychological Stress Evaluator (PSE)

and Voice Stress Analyzer (VSA) (Kubis, 1973; McGlone,

1975; Almeida et al., 1975). However, independent research

has shown, for example, that at least two of these devices

(the PSE and VSA) probably are not capable of the analysis

claimed by their proponents (Barland, 1973; Kubis, 1973;

McGlone, 1975). For example, the PSE reportedly evaluates

stress by demodulating the inaudible, stress related.FM

patterns known as "Muscle Micro Tromors (IMe) (Almeida et al.,

1975). However, Almeida et al. conclude "our results, how-

ever, do not confirm the theoretical basis of the PSE" (p. 4).

Speaker identification is another area within Forensic

Commuhication thlat has been shown to be affected by stress

(Hollien &c Majewski, 1977; Doherty & Hollien, 1978).

Hollien and Majewski found that the ability of the long-

term speech spectra (LTS) technique to identify individuals

was reduced somewhat under conditions of stress--induced

by randomly applying electric shock to the subjects while

they were speaking. When a fullband procedure (80-10,000 H~z)

was used, the effect of stress was to decrease the identifi-

cation scores by 8"/o over the unstressed (normal) speaking










condition. Moreover, this difference (between the normal

and stress identification scores) was greater (200/) when

the samples were bandpassed (315-3150 Hz) to simulate a

telephone transmission system. Therefore, it would appear

that the presence of stress not only degrades the LTS

speaker identification technique in a laboratory setting,

but when "in-the-field" restrictions are imposed, this

negative effect is magnified.

Doherty and H-ollien (1978) examined a set of three

speaker identification vectors in the presence of'speaker

and system distortions; these vectors consisted of long-

term speech spectra (LTS), speaking fundamental frequency

(SFF) and speaking time (ST). They used stress, induced

by electric'shock, as one of the speaker distortions; system

distortion was produced by subjecting the LTS vector to

transmission distortion--specifically, to a limited pass-

band of 315-3150 Hz. Doherty and H~ollien conclude that,

"while the described approach functioned adequately for

the normal speaking condition, no vector (singly or in com-

bination) adequately differentiated talkers when speech

was distorted" (p. 1).

It is assumed that, while committing a crime, the

criminal is under stress and his speech may be altered.











Therefore, the changes that occur in the criminal's speech

due to stress may influence the prob7ability of making a

correct identificat-ion. It would appear that knowledge

about the effects of stress on speech is important to the

area of Forensic Communication. Conversely, the development

of, and attempts to use devices such as the PSE indicate the

need for increased information about stress and its effect

on speech.


Diver Comnmunication


The relationship between stress and its effect on

speech is also important to Diver Communication. Once the

diver enters the water, he is in an alien environment that

is dangerous and potentially fatal; thus, he is ipso facto

in a stressful situation. Monitoring the diver's speech

in order to determine his emotional state would add another

safety feature to a diving operation. Moreover, knowledge

of a diver's specific emotional state should be useful in

anticipating and preventing diving accidents: in turn, a

safer working environment sh~ould result.

Safety measures are even more important in the case of

saturation diving. In addition to the dangers and stresses

of "shallow water" diving, the Jaturatted diver remains









submerged for long periods at great depths and is unable

to surface safely in case of an emergency. Furthermore,

a saturated diver tends to be assigned a heavier work

schedule and lives in the highly restricted confines of an

underwater habitat which leads to more stress--at least,

when the situation is compared to conventional diving.

Voice communication channels are usually available

during most diving operations. The ability to use existing

communications to monitor a diver's level of stress would

add to the safety of diving operations without imposing

additional burdens on the divers.


Aerospace Communications


'Ihe analysis of speech to determine emootional states

also is important to Aerospace Communication. The emo-

tional state of pilots and astronauts based on speech

samples has been investigated by authors such as Williams

and Stevens (1969), Simonov and*Frolov (1973) and Kuroda

et al. (1976). Williams and Stevens (1969) analyzed

speech samples obtained during obviously (emotionally)

stressful situations. Their recordings consisted of nir-

to-ground communications between pilots and control tower

operators during emergency (i.e., stressful) situations,










plus a recording of the radio anlnoulncer describing the

H~in~onburg disaster. These authors obtained fundamental

frequency (fo) contours of the stressed speech using narrow-

band sound spectrograms sampled at 0.15 see intervals. They

found differences in fo contours of the speech samples re-

corded during the emergency situation as compared to samples

before the emergency. The normal (nonstressed) samples

were characterized by smooth, slow, and continuous changes

in fundamental frequency as a function of time; the stress

contours exhibited fluctuations that often were neither'

smooth nor continuous. Williams and Stevens state that

when emotional stress is experienced, the speech of the

individual miayi exhibit "sudden changes or jumps in funda-

mental frequency from one syllable to the next, and rapid

up-and-down fluctuations may appear in the contour" (p. 1372).

Furthermore, they found an increase in the mean fo for the

stressful condition as compared to normal speech--and the

range of fo was greater for stress. Williams and Stevens

conclude that measurement of the median fo and range. of fo

may serve to classify whether an individual is undergoing

emotional stress, providing, of course, his normal values

are known.










Simonov and Frolov (1973) estimated the emotional

stress experienced by two cosmonauts during several phases

of the flight of "Voskhod 2" and during training in heat

and pressure chambers. They used a one-third octave spec-

tral analysis within the range of the first formant (300-

1200 Hz) and compared these results to measures of heart

rate (beats/minute). These authors were able to diffor-

entiate the degree of emotional stress in their subjects

about 85"/o of the time by means of this method.

.Kuroda et al.(1976) analyzed the communications of

pilots in 14 actual aircraft accidents, eight of them fatal,

using a measure of fundamental frequency; specifically,

the vibration space shift rate (VSSR). VSSR is calculated

by analyzing the spacing between the vertical striations

on wideband sound spectrograms; specifically, the widest

spacing between vertical striations during the normal phase

of a flight is compared to the widest spacing encountered

during an emergency situation. They found that the highest

VSSR occurred during the initial phase of an emnergency

situation. Furthermore, when the emergency resulted in a

fatality, Kuroda et al. found that a high VSSR had been

main-tained during the emergency. Therefore, Kuroda et al.

(1976) conclude that "by analyzing the voice communications










of a pilot involved in an emergency situation, we have a

method of determining whether stress itself could have

been a contributing factor in thle outcome of that in-

flight emergency" (p. 533).

As can be seen from the above discussion, research

on the manifestation of stress via speech is indeed appli-

cable to numerous areas of speech communication. Its

application ranges from assisting law enforcement agencies

in their investigations to avoiding a possible fatality in

diving and/or aerospace operations. However, the research

conducted to date has been limited in its scope, leaving

many questions unanswered.


A Definition of Stress


If research is to be carried out on stress, it would

appear that the first step would be to operationally define

this psychological state. Many attempts at such a de-

finition have been reported. For example, in Psychologi-

cal Stress: Issues in Research, Appley and Trumbull (1967)

state, "It is further evident from the definitions cited

that another area in which separation of psychological

from physical aspects is required is that of threat" (p. 36).

They also quote E. A. Haggard who stated, "An individual










experiences: emotional stress when his overall adjust-

ment is threatened, when his adaptive mechanisms are

severely taxed and tend to collapse" (p. 39). Therefore, it

would appear that threat is a major variable in determining

emotional (psychological) stress--at least of certain types.

That is, an individual must perceive a threat to his ego,

integrity, values or goals. Lazarus (1966) states that the

individual must "anticipate a confrontation with a harmful

condition of some sort" (p. 25); he goes on to say that

"the strength of the stress response is determined by the

strength of the intervening process, that is, by the degree

of threat" (p. 26). Furthermore, emotional stress can be

defined as a reaction to a stimulus condition; for example,

Basowitz et al. (1955) concluded, "We should not consider

stress as 'imposed' upon the organism, but as its 'response'

to internal or external processes which reach those threshold

levels that strain its physiological and psychological in-

tegrative capacities close to or beyond their limits" (pp.

288-289). Further, Appley and Trumbull (1967) suggest that

stress "is a response state and that its induction depends on

the mediation of some appraising, perceived, or interpreting

mechanism" (p. 46). Therefore, it would appear that, in part,

emotional stress is a mediated reaction to a threat to an

individual.










Often the terms stress and emotion are used inter-

changeable, but, are they synonymous? A given stressful

situation may produce dissimilar emotions in different

people as well as diverse emotions in the same individual

at different times. Some emotions (grief, fear, anger) are

invariably connected with stress, but that does not neces-

sarily equate stress and emotion. "There are many emotions

that have nothing to do with stress (love, joy, delight)"

(Arnold, 1967, p. 50). Arnold has called the emotions

accompanying psychological stress "contending emotions";

they are primarily anger and fear and their combinations.

Therefore, as stated by Levitt (1967), stress "appears to be

a kind of operator word which is applied in connection with

emotion-evoking situations and reactions" (p. 15). Based on

the above discussion, it would appear that a reasonable

definition of stress is that it is a psychological state that

is a respns to a perceived threat and it will be accom-

panied by specific emotions.


Laboratory Induced Stress


Stress can be operationally defined in the laboratory;

that is, as long as the definition meets the above criteria.

Further, the definition can be based on, or related to, the










use of a particular stressor of either physiological or

psychological origin. For example, physiological stressors

are usually employed to induce physical discomforts; they

include electric shock (Forr & Seaver, 1975; Doherty &

Hollien, 1978; Hollien & Majewski, 1977; Silverman &

Silverman, 1975) as well as olfactory stressors (Ostwald,

1963; Farr & Seaver, 1975). .On the other hand, procedures

to induce psychological stress include having subjects (1)

give a five-minute speech to a group of peers or (2) sit

in a small room for ten minutes w~ith the object they are

afraid of most (Farr & Seaver, 1975). Finally, it is im-

portant that the laboratory task be realistic compared to

the task the subject will ultimately have to perform under-

real stress (as distinguished from laboratory stress). In

this case, the investigator's conclusions of subjects' be-

havior-under real stress will be more valid.


Situational Stress


In this case, the experimenter ha~s niot induced the

stress; rather, the situation is normally stressful to the

subjects. In situational stress, the consequences often are

more severe than, for example, they would be if the subjects










received a mild electrical shock. In this case, their

lives might be threatened--as in the case of pilots or

astronauts in emergency situations (Kuroda et a., 1976;

Simonov & Frolov, 1973 and 1977; Williams & Stevens, 1969).

In addition, the subjects may not have the option of avoid-

ing the stressor. HIowever, situational stress need not be

life threatening; for example, students taking an oral

examination or giving a public address might experience a

high level of stress even though their life is not in danger.

In this case the stress is of a psychological nature since

the subjects are not being physically threatened.

Therefore, the investigator can either create and

induce stress in the laboratory or utilize the stress pro-

duced by a specific situation. In addition, he can use any

of several measures to describe the effects of the stress.

These measures can range from physiological to subjective/

perceptual.


Measures of Stress



Physiological M measures


The external manifestation of emotions has been of

interest to researchers for many years. For example, Charles










Darwin (1872) discussed the use of facial and body move-

ments as well as postures in his book The Expression of the

Emotions in Man and Animals. However, Darwin's text is only

descriptive in nature. Various methods subsequently have

been used in an effort to measure the physiological correlates

of emotional stress; these include (1) heart rate (Deane,

1961; Bowers, 1971; Rankart & Elliot, 1974); (2) galvanic

skin response (G;SR) (Bowers, 1971; Bankart & Elliot, 1974);

(3) electroencephalogram (EEG) (Itil et al., 1976); (4)

respiration (Sovijarvi, 1974); and (5) blood pressure (Brod

et al., 1959; Blair et al., .1959).

While these measures have resulted in relatively

accurate Indlcators of stress, thery c~an Iin~roducel prob~i`leS

.when used "in-the-field." For example, physical contact

with the subjects is required to obtain any of these measures.

Such contact does not create serious problems in clinical-

or laboratory applications but prolonged monitoring (o-f

astronauts or divers,tor example) can result in severe

complications. Despite attempts to eliminate the problem,

such as design improvements in the sensors and the use of

creams or jellies'at the placement site, extended attach-

ment of sensors may lead to skin irritations and other

undesirable consequences. For this reason, it has been










recommended that GSR, respiratory rate, EEG and other

physiological measures be recorded only intermittently

and the sensors removed after each recording. In addition,

subjects' physical activity is restricted while the sensors

are attached and he "becomes an observed subject who is by

no means indifferent to the experiment in which he is in-

volved" (Simonov & Frolov, 1977,-p. 23)--a condition that

possibly can affect the resultant measures.

Furthermore, attachment of electrodes or measuring

equipment tends to distress a patient under psychiatric

treatment. However, the therapist could easily tape record

the therapy sessions as a standard procedure. The use of

a tape recorder during therapy would disrupt the session

minimally. The patient would simply accept the recording

process as part of the psychiatric environment. Tape re-

cording the therapy session would provide a permanent

account of the session for later evaluation by the thera-

pist. In addition, the recordings could be used as feedback

for the patient, which can be important in psychiatric

treatment.










Acoustical/Tomporal
Measures of Stress



Analysis of the speech signal as an index of emotional

stress is becoming a viable approach to research due to ad-

vances in the state-of-the-science of acoustic phonetic

analysis equipment and techniques. However, the parameters

of the speech signal, which are critical to the identification

of stress, as yet have not been determined. Moreover, pre-

vious investigations have, focused primarily on the frequency

domain of these signals (e.g.,- fundamental frequency, range

of fundamental frequency and formant structure) and have

neglected intensity and temporal parameters. It would

appear necessary that the parameters~otr each of these ljidomins

be investigated as possible indicators of emotional stress.






Although-it is a totally different type of stress,

the study of research on linguistic' stress (emphasis) can

be of aid to the Acoustic Phonetician. Linguistic stress

has been found to influence the'intensity of a stressed

word. In a 1937 study by Tiffin and Steer, it was found

that the average intensity of the stressed production of a

word was 9.3 dB higher when compared to its unstressed










production. Further, the stressed words were more intense

than the unstressed about 71% of the time. On the syllabic

level, Ortleb (1937) found that when compared to unstressed

syllables, the emphasized syllables were more intense in

both emotional and unemotional material; however, the in-

tensity increase was more pronounced in the emotional ma-

terial. Further, Schramm (1937) observed that the intensity

of accented syllables was greater than the unaccented sylla-

bles in 78% of the words studied. Therefore, it would

appear that intensity changes are important indicators of

linguistic stress.

When intensity features related to emotional stress

have been-studied, the relationship between intensity and

stress has not been found to be obvious. For example, when

studying emotional stress, Hecker et al. (1968) did not

find consistent intensity differences between their control

(nonstress) and stress conditions. Six of their ten sub-

jects exhibited small and inconsistent differences in in-

tensity between the two conditions. Of the remaining four

subjects, only one showed an increased intensity for the

stress condition while the remaining three subjects lowere

their intensity level under stress. In a study by Friedhoff

et al. (1964), subjects were instructed to respond no to a









series of questions even though the correct response to

some of the questions should have been yes; this technique

was utilized to establish a mildly stressful lying situation.

These authors report that average intensity increased under

the condition of lying; however, variability was large,

i.e., "some individuals give a consistently louder and some

a consistently softer response when lying." Friedhoff et al.

also measured the intensity of the response two-six-nine

under normal (nonstress) and stressed conditions, using

.electric shock as the stressor. They found that.maximumm

intensity shifted from the last word (nine) for the normal

condition to the first word (two) for the stress condition.

However, it is not.clear from their study whether they are

reporting a trend across subjects or the results from only

one subject. Therefore, it would seem that the manner in

which intensity manifests emotional stress is not as well

defined as it is for linguistic stress. However, intensity

variability and the amount of intensity change might prove

to be important indicators of emotional stress.


Fundalmental Frequency (fo)


The fo of speech also may be a parameter which is in-

fluenced by stress. For example, Fairbanks and Provonost










(1939) studied the pitch characteristics of sixe actors,

reading the same test passage and expressing five emotions:

contempt, anger, fear, grief, and indifference. They

measured (1) pitch contours,1 (2) pitch level, (3) pitch

range, (4) rate of pitch change, (5) extent of pitch shifts,

and (6) the number of changes in the direction of pitch

movement per second. These authors report that they could

distinguish among the five emotions they studied solely on

the basis of the fo measures. For example, they found that

fear was identified as having the (1) highest median pitch

level, (2) widest total pitch range, (3) largest excursion

of all pitch shiifts (together with anger) and of upward

shifts within phrases, (4) largest change in pitch level

between phrases,' (5) fewest pauses during which no pitch

change occurred, and (6) highest number of pitch changes per

second of speaking time. Therefore, it would appear that

fundamental frequency characteristics serve to differen-

tiate among emotions.

In aIn investigation of the affects of task-induced

stress on speech, H~ecker at aIl. (1968i) required ten sub-

jacts to road six meters and report the sum of the readings

together wlith a test phrase. In this experiment they could


1Fairbanks and Provnost used the term pitch inter-
changeably with fo.










vary the level of stress induced in the subjects by con-

trolling the duration of the meter display. The results

of this study indicated inconsistent changes in fo across

subjects. That is, some of the subjects raised their

fundamental frequency while others lowered it as a

function of increased stress. These authors point out

that those subjects who lowered their fundamental frequency

also spoke at a lower intensity level--as apparently these

parameters were interrelated.

Somewhat different results have been reported by

Williams and Stevens (1969) who conducted their study

using .the air-to-ground communications between pilots and

control tower operators during known stressful (i.e.,

emergency) situations. In an analysis of both pilots'

and control tower operators' voices (for indications of

stress), they obtained the level and range of f, and fo

contours (from narrow band spectrograms). They found that

under increased levels of stress the pilots and air traffic

controllers raised their fo, and showedl "irrcqularities

and discontinuities" in f, contours. Williams and Stevens

(1969) also studied fo usage in the voice of the announcer

who described the H~indenburg zeppelin crash--both before

and after this disaster. In addition to an increase in fo










after the crash, they found that the announcer greatly

reduced his inflectional pattern (variations in fo) during

the period he was under high stress.

In 1972 Williams and Stevens investigated the acoustical

correlates of several emotions (anger, fear and sorrow) as

portrayed by actors. In their analysis of these simulated

emotions, they observed differences in mean fo and range of

fo. They conclude that measurement of the median fo and

range of fo for a speech sample serves to differentiate

among emotions--assuin that the-normal fo and range of

f, for the talker are known.

On the basis of these experiments, it would appear

that~ a speaker's emotional state can be determined by

analyzing the acoustic speech signal. However, it can be

seen that disagreement remains as to the affect of stress

on fo as Hecker et al. (1968) report intersubject~vari-

ability while Williams and Stevens (1972) indicate that fo

is raised when subjects undergo emotional stress. Some

discrepancies also can be found in the data contrasting

emotions based on fo measures (Fairbanks & Provnost,.

1939; Williams & Stevens, 1972). Finally, when reporting

changes in fo, authors often describe: irregularities in

to (rapid fluctuations) or irregular "bumps" in fo









contours, "irregular bumps in the contour, which might

be interpreted as a kind of tremor" (Williams & Stevens,

1972) and "vibration of the vocal folds was too irregular"

(Hecker et al., 1968).Whnethe statements above are descrip-

tive and observational in nature, analysis of these rapid

fluctuations in fo may prove useful in the detection of

emotional stress. Thus, it would appear that the relation-

ship between stress and its affect on speaking fundamental

frequency requires further investigation.


Temporal Measures


The temporal characteristics of speech have also been

suggested as potential Inaicators of nothl em~otional and

linguistic stress. For example, increased duration of

both words and syllables was found to be a major correlate

of linguistic stress (Ortleb, 1937; Schramm, 1937; Tiffin

& Steer, 1937). Further, Fairbanks and Hoaglin (1941)

found that simulated emotions (anger, fear, indifference,

contempt and grief) could be differentiated to some ex-

tent on the basis of speaking rate, speech to pause-time

ratio, and durations of speech and pauses. However, they

report that the five emotions could be divided into two

groups based on these temporal measures, with anger, fear,










and indifference in one group and contempt and grief in

the other. All three of the first group of emotions are

characterized by rapid speaking rate and shortened dura-

tions of phonation and pauses, but did not differ signifi-

cantly from each other. Contempt and grief, on the other

hand, could be differentiated both from each other and from

the other emotions. Although they were both.characterized

by a slow rate, the rates (ini simulating contempt) are

produced by increasing the duration of both speech seg-

ments and pauses. On the other hand, the slow rate for

grief results almost entirely from prolonged pauses,

particularly those between phrases.


P.isflagcnc


An additional parameter that may serve as an indicator

of stress~ is speech fluency/disfluency. For example,

Silverman and Silverman (1975) have defined this charac-

teristic (disfluency) as follows: "repeating part of a

word or an entire word, repeating a p~hrase, inserting

extra sounds between words (ah, u~m), changing the wording

of a sentence, prolonging the sounds of a word, or breaking-

uip the sounds of a1 word." In their study, Silverman and

Silverman informed their subjects that an electric shock









would be administered for each instance of disfluency

during the reading of a 330 word passage. They found

that under thiis stressful condition, normal speakers

tended to become more fluent. Thiat is, under stress the

normal speakers were more accurate in their speech produc-

tion as compared to the nonstress speaking condition. In

short, it would appear that this feature may serve as an

indicator of a speaker's emotional state.



Subjective Judgements
of Stress


Stress can be quantified in other'ways. These ap-

proaches include subjective measures such as aural/percep-

tual identification of .stress by listeners and self-ratings

by the subjects themselves of the level of stress they ex-

perienced.


Aural perception


It is generally agreed that the stress an individual

experiences can often be detected merely by listening to

his speech. Although nural/perceptual procedures are not

as objective and analytical as the physiological, acousti-

cal and temporal measures previously discussed, they remain










valid dimensions for determining a speakers emotional

state. For example, Fairbanks and Provnost (1939) ob-

tained correct aural perceptual identification of five

emotions ranging from 66% to n88%. Therefore, they con-

clude that emotionss expressed by the voice alone are

readily identifiable" (p. 104). Furthermore, Hiecker et al.

(1968) used an aural/perceptual procedure in their study

and found that listeners were able to correctly identify

stressful responses from 66% to 83%/ of the time.

An aural/perceptual evaluation is of major importance

in any study using simulated emotions as speech material'

in order to verify the subjects' accuracy in affecting the

emotions under investigation. Furthermore, the acoustical/

temporal parameters critical to the perception of stress

could be determined by comparing speech samples judged to

be normal and those produced under stress.


Self-rating by subjects


In addition to the measures previously cited (physio-

logical, acoustical, temporal and aural/perceptual), self-

ratings by the subjects themselves are valuable in determin-

ing the level of stress they have experienced (Acker &c Rey-

nolds, 19366; Zuckerman at al., 1964). Inldeed, this technique









can be utilized to obtain an index of both levels of

stress and possible intersub~ject differences in l~evel.

That is, what may be stressful to one subject may not stress

another individual. Merely assuming that all the subjects

had experienced a significant (or equal) level of stress

could lead to erroneous conclusions based on the obtained

data. For example, Farr and Seaver (1975) found that sub-

jects rated various experimental protocols differently-

based on the ~level of stress produced. Therefore, self-

rating by the subjects is -important in order to insure that

the subjects perceived a stress producing threat and to be

able to quantify the level of perceived stress.


Objectives


The thrust of the current study is focused on an acous-

tical/temporal analysis of the effects of stress on speech.

Included in the research protocol were both laboratory

induced stress and nonlaboratory, or situational, stress.

The laboratory stress was induced using a mild electric

shock as the stressor. The situational stress condition

consisted of speeches given by students enrolled in a

university level public speaking course.









It is possible that, with further research, the amo-

tional and stress states of an individual could be deter-

mined by an analysis of speech features. With this objec-

tive in mind, the current study consisted of an analysis

of several acoustical and temporal speech parameters produced

under emotional stress. Included in the acoustical analyses

are intensity and fundamental frequency. Furthermore,

several characteristics of each acoustical parameter weree

analyzed. For example, the following characteristics within

the intensity domain were measured: (1) maximum intensity,

(2) mean intensity, (3) the mode of the intensity distribu-

tion, and (4) the intensity distribution itself. Similarly,

seVeral Characiteristics WIthinI the~ temporall domaII ~ in la

were analyzed. The temporal characteristics included.

(1) speech/pause ratio, (2) speech rate, (3) the time-

energy distribution, (4) the number of speech bursts,- (5)

the number of pauses, and (6) speech time/total time ratio.

In addition, the number of disfluencies in the two-minute

speech sample also were measured.

A comp~irison w~as made between normall speech andit apeech

produced under stress. The comparison of these two speaking

conditions was based on the acou~sticall/Lmompora parameters

analyzed. The ultimate goal of the present study is to










add to the corpus of knowledge concerning~ thec offects of

stress on speech.

However, the specific goals of this research may be

stated in the form of several questions. They are as

follows :

1. Which of the many selected acoustical/temporal
parameters,. found within the speech signal,
might serve as indicators of emotional stress?

2. Will stressful speaking conditions signifi-
cantly increase or decrease the levels of the
measured parameters?

3. Is the resultant change due to stress in a
specific parameter consistent for all speakers?















CHAPTER II


METHOD


The objective of this investigation is to compare

speech produced under.stress to nonstressed (normal)

speech. The analyses carried out included both acoustical

and temporal measures as well as the level of disfluency

observed and a self-rating (by the subjects) of the stress

they experienced. The acoustical analysis consisted of

several measures based on (1) speech intensity (SI) and

(2) speaking fundamental frequency (SFF). The temporal

measures analyzed were (1) speech/pause ratio (SPR),

(2) speech rate (SR), (3) the time-energy distribution

(TED), (4) the number of speech bursts, (5) the number of

pauses, and (6) speech time/total time ratio (ST/TT). In

addition, the number of disfluencies was determined for

both speaking conditions (stress and nonstress); two

minute speech samples were used for all the analyses.

The self-rating by the subjects obviously was based on

each subject's overall reaction to the experimental pro-

cedure and was a criterion for subject inclusion.









The analyses utilized were common to two experiments,

i.e., stress induced by either (1) electroshock or (2)

public speaking was examined. Therefore, the types of

analysis carried out will be discussed first; these

descriptions will be followed by reviews of the procedure

used for each experiment.


Acoustical Analyses


Included among the acoustical analyses were several

parameters within the domains of speech intensity and

speaking fundamental f'requency. They are as follows.


Speech Intensity (SI)


First among the intensity characteristics to be

analyzed was mean intensity (MI). In addition to the

mean, the maximum intensity (MAKI) also was analyzed.

RAXI is defined as the highest intensity produced by each

subject for the entire two minute speech sample. The mode

of the intensity distribution (MODI) defined as the in-

tensity level most frequently produced, also was deter-

mined for each speech sample. Additionally, the entire

intensity distribution (ID) for each speech condition was

analyzed for differences between the normal and stress

speech samples.









A computer aided procedure wa7s used to obtain th~e

various intensity measures; a Digital Equipment Corporation

(DEC) PDP-8i minicomputer and analog-digital (A/D) converter

were used for this purpose. First, the envelope of the

speech signal was obtained using a rectifier/integrator

circuit. The speech envelope was negative do shifted in

order to make full use of the A/D range (-1.0 to 1.0 V).

The computer software (ADCNV), controlling the A/D con-

verter, allowed the computer operator to specify the parame-

ters of sampling rate and the length of the speech sample.

The sampling rate used was 50 samples per second and the

sample length was 120 sec (2 min). This rate was deter-

mined by te~scing Iusierous~ d~jiT~fferen iiaLl inl orertoob

tain the lowest sampling rate that would accuratelyy repre-

sent the speech envelope. The output of the A/D converter

was then stored on magnetic tape (DECTAPE) and the analysis

program (INTANA) was implemented. INTANA was written to

give the operator substantial control over several program

parameters. The first parameter the program requires the

operator to enter is the A/D value of the dc shift (AD

ZERO) used during the inputting of the data. Next, the

operator specifies th~e A/D value of a de calibration voltage

(AD CAL). This calibration value is used by INTANA to










convert: A/D units to dB values of intensity. The next

parameter requested by INTANA defines the calibration vol-

tage as a multiple (MULT) of the dD reference level selected

by the operator. This input was followed by a dB adjustment

(DB ADJ) used to compensate for any amplification or attenua-

tion of the original signal. That is, if the signal had been

amplified at the time it was recorded, the operator would

onter the negative value of the amplification lovel (in dB).

If, on the other hand, the signal had been attenuated, the

value of DB ADJ'would-be positive--in order to add back the

amount of the original attenuation. Following the DB ADJ, the

program requested a dB cutoff (DB CUTOFF). Any signal level

below DB CUTOFF was considered to be either noise or a pause

and was not included in the computation of the various in-

tensity measures. The intensity analysis was then begun

using the A/D equivalent of DB CUTOFF as well as the ~previous

operator specified parameters.

The INTANA output included the A/D values of the minimum

(AD MIN) aInd maximum (103 MAX) Icvols of the stored data, as

well as thle actual number of sample points for AD M1IN and. AD

MAX and the corresponding percentage of the total data for

each. The total number of valid (i.e., speech) and rejected

(i.e., noise or pauses) data points and. their corresponding










percentages wlere indicated by thie output of INTLANA as were

the maximum (1DB MAX), range (RANGE), mean (DB MEAN), and

standard deviation (STD DEV) of intensity (in dB3). Next, a

distribution table of intensity levels was printed; it

listed the dB value, the number of sample points at each

level, and the corresponding percentages. The final output

was a histogram of the intensity values.

In summary, the intensity measures used are the mean

(MI), maximum (tS\XI), the mode of the distribution (MODI),

and the overall distribution (ID). As stated, these inten-

sity measures were utilized, in comparing speech produced

under stress and nonstross conditions. The rationale for

using so many intensity .measures is that it remains to be

determined which speech intensity measures are critical to

detecting the presence of stress. Therefore, it was judged

that as many intensity parameters as possible should be in-

vestigated and the critical measures identified.


Speaking Fundamental
Frequency (SFF)


Speakcing fundamental freqluency (:;FF) in a1 m~nsure of

the rate of opening and closing of the vocal folds. In the

present study the SFF measures of interest, i.e., mean SFF









and the SFF distribution, were compared as a function of

the stress and normal speaking conditions.

The specified SFF measures were obtained using the

Fundamental Frequency Indicator (FFI) available at the

Institute for Advanced Study of the Communication Processes

(IASCP). FFI is a digital fundamental frequency tracking

device consisting of low-pass filters with cutoff at half-

octave intervals coupled with high-speed switching circuits.

FFI produces a series of pulses, coinciding with the funda-

mental period of a complex speech waveform. These pulses

are then processed by a DEC PDP-8i minicomputer using an

internal clock to mark the time interval between pulses and

this data is processed to yield the geometrical mean trequency

and standard deviation of' the fundamental 'frequency distribu-

tion. In addition, FFI produces a SFF distribution table

and a plot of the SFF distribution. ~The distribution table

produced by FFI lists the following data: SFF in (1) semi-

tone (ST), (2) absolute frequency (Ilz), (3) the number of

occurrences of each value of SFF, and (4) the per cent of the

total phonation time each measure OE SFF was produced. The

value of SF~F in H~z and STr and thre per cent of occurrence were

the values used from the SFF distribution table.










The mean SFF, both within and across sexes, was com-

pared between speech samples produced under normal and

stressed conditions. In addition,. the SFF distribution

for each subject was used to compute a composite SFF dis-

tribution for each sex. That is, a composite distribution

was determined for the males and one for the females and

each used to compare the two speaking conditions within

sexes.



Temporal Analyses


In addition to the acoustical parameters discussed

above, temporal aspects of speech also were investigated as

possible indicators of emotIional stress.- The temporal

characteristics analyzed consisted of (1) speech rate

(SR), (2) a time-energy distribution (TED), (3) speech/

pause ratio (SPR), (4) the number of speech bursts, (5)

the number of pauses, and (6) speech time/total time ratio

(ST/TT).






Speech rate (S]R) is defined as the number of syllables

produced per second. This temporal future was measured at

the syllabic level due to varying word lengths; obviously,










polysyllabic words will take longer to produce than mono-

syllabic words. Therefore, using word boundaries would re-

sult in misleadling data. The syllabic level of measurement

will become most important for the stress speech samples

obtained in the public speaking experiment (to be discussed

later) because the intersubject speech samples will be

different; i.e., they are not the same samples.

The obtained two minute samples were divided into 15 sec

segments and SR calculated for each segment by dividing the

number of syllables in each segment by 15 sec. Finally, the

mean and standard deviation of SR for the two minute samples

were calculated.


Time-Energyi Dis tribution (TED)


The timne-energy distribution (TED) paramneter originally

was developed in 1978 by Johnson for a speaker identification

study. "In general terms,. TED reflects the total time a

talker's speech intensity remains at a specific energy level

relativee to hiis peak amplitude). It also provides indica-

tion of the speaker's speech pattern with respect to speech

bursts and pause periods" (Johnson, 1978, p. 39). In the

present study, TE~D was added to the other measurements in

an attempt: to discriminate between stress and normal speech.










TED was obtained utilizing a PDP-8i minicomputer and

A/D converter. Initially, the envelope of the speech sample

was generated using a rectifier/integrator circuit. The

speech envelope then was digitized (A/D) and analyzed in

real time using software developed by Johnson (1978). This

software segments the speech signal into ten equal intensity

levels relative to the maximum intensity produced by the

speaker. Based on this segmentation, thle TED program com-

putes : (1) the-numbe~r of speech bursts, (2) speech bursts

per sec, (3) time in sec, (4)- per cent of the total time,

(5) mean time in millisec, and (6) standard deviation in

millisec of each speech burst for each of the ten intensity

levels. .The mean pause periods and the number of pauses

were the direct reciprocals of the speech bursts. In this

manner, TED determines the distribution of both the speech

bursts as well as the pause periods. The speech burst dis-

tribution was used in order to analyze differences between

speech samples produced under the two speaking conditions

(stress and no stress). In addition, the TED output

also was used to determine the speech/pause ratio (SPR),

the number of speech bursts, the number of pauses, and the

speech time/total time (ST/TT) ratio (see below).









Speech/Pause! Ratio (SPR)


The speech/pause ratio (S"R) is the ratio of the per

cent of speaking time to the pr cent of pause timue. A

pause is defined as that period of time during which an

individual is not producing any audible sound. SPR can be

readily obtained using the TED printout; that is, SPR can

be calculated by using the percent of speech time and pauses

for intensity level one, th~e lowest level.

Each two minute speech sample was divided into eight

15 soc segments and SPR was calculated for every other seg-

ment beginning with the first segment. This approach allowed

the investigation of those SPR changes within each sample

that may have been "averaged out" by examining the entire

two minute sample. Furthermore, this segmentation permitted

the calculation of the mean and standard deviation of SPR

for the entire sample.


Speech Bursts and Pauses


In aIddition to SPR, specific speech bursts and paulses

were counted in order to determine possible differences

between normal (nonstress) speech and speech produced under

stress. The rationale for this analysis was as follows:'c a











reliance only on SPR would result in certain confusions.

That is, one speech burst of 20 sec and four bursts of

five sec cach would both result in a SPR of' 0.5 for a 40

sec sample, when in fact, the two samples are different.

The number of speech bursts and pauses were obtained from

the lowest TED intensity level. This intensity level

distinguishes between portions of the total sample that

include vocalizations and those that do not (i. e.,

pauses).



Spechc~ Time/Trotal Time_



The speech time/total time ratio (ST/TT) is the ratio

of thre amount of time in which speech occurred to the total

time of the speech sample. This measure compares the

speech time of a given sample to the total time of that

sample. The ST/TT ratio was calculated utilizing the TED

output. Both the ST and TT of each speech sample were

obtained from level one of th~e TED.

The same segmentation of the total simple used to

calculate SPR and SR applies to ST~/TT. That is, eaich










sample was divided into eight 15 sec sub-samples and

starting with the first sub-sample, alternate sub-samples

were used to calculate the mean and standard deviation

of ST/TT.


Disfluency


The final parameter analyzed w\as speech disfluency.

This parameter is neither an acoustical measure nor

entirely a temporal measure. While disfluency does relate

to the temporal domain, in the present study it was not

measured as a function of time. Instead, each occurrence

of a disfluency was merely counted. Therefore, this

parameter will be discussed separately from the acoustical

and temporal parameters.

The definition of disfluency used in the present study

is: "repeating part of a word or an entire word, repeating

a phrase, inserting extra sounds between words (ah, um),

changing the wording of a sentence, prolonging the sounds

of a~ wordl, or breaking-up thc s~ounds; of a wordi" (Silverma7n,

& Silverman, 1975, p. 353). The present author and a

speech pathologist listenedl to each two minute sample and











counted each time the speaker was disfluent. Included

among the types of disfluencies measured were "audible

pauses"; that is, sounds iniserted between words (e.g., ah

and um). These audible pauses would have been measured~ as

speech in the SPR measure discussed above. Inclusion of

audible pauses in; the SPR would have artific-ially inreased

the SPR. This possibility would have been obscured with-

out separately examining disfluencies. In ;Iddition,

although the experimental paradigm was different fr~om ^Ihe

present study, in their study Silverman and Silverman found

that disfl~uency and the threat of electric shock are

correlated. Therefore, it would seem that the number of

disfluencies may serve as an indicator of stress.


Subjects' Self-Rating Stress


The individuals in both experiments rated the amount

of stress they experienced in order to select subjects for

inclusion in each experiment. TIhis celf-rating was acco:m-

plished u:sing the Multiple Affect Adjecctive Check List

(MA\ACL) (Zuickerman chl al., 196(4). Th'ie MAA~CLJ hasl th~rce

scales in border to measure: (1) anxiety, (2) hostility,










and (3) depression. However, only the anxiety score was

used for subject selection in the present study. The MA~ACL

has a maximum anxiety score of 21, with normal (nons tressed)

subject groups scoring 4-7 and individuals under stress

averaging scores of about 15. Each subject completed the

MIAACL response form prior to the initiation of the speaking

task under the stress condition. Only those individuals with

scores of 13 or greater were included in the experimental

population in order to insure that subjects had achieved

a significant level of stress. A sample of the MAACL may be

found in Appendix A.


Recording Procedure


The recording procedure will be discussed separately

from the experimental descriptions since it was common to

both experiments. Specifically, a FM wireless microphone/

transmitter was used to record the speech -samples as can be

seen in Figure 1. A special headset was designed and fabri-

cated in order to position the microphone in constant re-

lationshlip to thie speaker. This headset with the microphone

in place can be seen in Figure 2. Th~e rationale for con-

trolling the relationship between the microphone and the

subject was to eliminate intensity variations resulting from




























'I 1 7r I ifi ~ 'j~j 'l l I I i I i' '''' lll~ '' I'' l'''' i' \ll'''' i' il' j'qi i'

wo- so n R3 ;WEiSTCOTTQ ~~~x~ibl ~ ~S~tali n les s Stee RucI t or

lith ill id it lidl llrll I I lill l ll Hi l I 11 I II I I 11 1 il I n il 101 Ilithildil bl lith1 h nihitt ll d unhn inh


Figure 1. FM Wireless Microphonec/Transmiircer Used in Recording Process.


























:~ T~aplx ~llrri

~~t"J -'~ ,t" ?_ ~I- ;~~~g~
o Q~2~?p:';l
~15~: ~-~~I:I a I
iLbYi~; )1
(I
,~- -- --- -- G

. ~ ; ,r~ ..~.~r~'~J-ur;in~FII"

L ~
.'.-'rt.. :
,,
.\.. i.i,


F~igure~ 2. Spac Lal~ly Docs iglnedi Headrscet usdc~ to Hloldi IM
Miicrophone/Tralnsmi tter~ (ShownI~ in Plance) in
a7 Fixed Position Rielative to Speaker.









head movement. The speech signal from thei~ microphone/

transmitter was received on a FM tuner which was coupled

to a Sony TC-353D tape deck via a flewlett-Packard 350D

variable attenuator. The output of the FMI tuner and the

input to the tape dock were maintained at fixod levels such

that attenuation was required in oirdo>. La~ not overdrive the

input to the tape deck. Therefore, the recording level was

controlled w~ith the attenuator, which facilitated main-

tenance of a1 constant gain in the recording system for both

the normal and stress speech conditions for a given subject.

Maintaining a constant gain between speaking conditions

perm~itted intensity comparisons between conditions. Addi-

tionally, this recordings techniqueP was~ chose inorert

minimize th-e affect of the recording procedure on a subject's

level of s tress--particularly in the situation stress experi-

ment. Furthermore, it was judged that_ i; all~owed the sub-

jects freedom of movement not possible w-cil1 conventional

recording techniques while maintaining w::jrim~ental control.


Experiment I: Labora7torry Stre-cs;


inl a laboratory settling the experimentel r has~ grerater

control over the stressing environment ais comp~aredi to "real"

or situational stress. Furthermore, it wJould seem reasonable










to first test a new procedure under highly controlled con-

ditions before advancing to nonlaboratory, in-the-field

environments. Therefore, the first experiment in this study

was conducted in a laboratory environment.

Laboratory stress, as used in this experiment, is

defined as a deliberately contrived stress induced in a

laboratory environment. Ini this particular case an electric

shockc was used as the stressor.






The subjects utilized in this experiment were recruited

from members of the Institute for Advanced Study of the

Communication Processes (IASCP) and the Speech Department

of the University of Florida. They ranged in age from 20 to

31 years, with a mean age of 23.9 years. The total popula-

tion included 7 males and 5 females; none exhibited any

speech/voice abnormalities. All subjects scored 13 or

above' on the MAACL anxiety scale under the stress condition.


Speech M;atorialls


Speech materials consistedt of two readings of a

modernization of R. L. Stevenson's ";ul Apology for Icalerj."

First the subjects read the passage with no electric shock









administered. (i.e., no stress induced). This reading con-

stituted their normal speech sample. Next, the subjects

randomly received an electric shock while reading the stan-

dard passage, and this was used as their stressed speech

sample. "An Apology for Idlers" was selected as reading

material based on the time required to read the passage;

approximately 3-3.5 minutes. A passage of this length was

needed in order to extract thle two minute speech sample

used in this experiment (se~e Appendix C).



Experimental Procedure


Recordings for this experiment were obtained by

utiliZaltionl of thei pro~cedurles descr~;~ibe ;labove sub~jecss

were recorded in an Industrial Acoustics Company (IAC)

sound treated room. Only the FM~ receiver was placed in the

IAC room with the subject, while the remainder of the re-

cording equipment was operated outside the room and

coupled to the receiver via the IAC room's patch panel.

The subjects' first task was to read "An Apology for

Idlers" with no stress being induced (i.e., no shock was ad-

ministerod) ini order to obtain a normal speech sample. At

this juncture, the electroshock procedure was explained and

the cloctrades were placed on the index and ring fingers of










the subject's hand. Once the electrodes~~ were in plaice,

the subject was asked to sign a Subject Informed Consent

Form (see Appendix B3) and complete the MA~ACL test.

The subject was told that the number of shocks (one to

seven) and when they would occuJr had been randomly assigned,

but that at least one shock would, be administered. In fact,

all subjcts received seven shocks given on the same pre-

selected words during the readingy.

A Grason-Stadler Psychogalvanometer, Model 4, constant

current sh-ocking device was used to administer the electrical

stressor. A current level of 2.5 ma was delivered manually.

Of course, this procedure did. not harm the subjects; it only

caused enough discomfort and threat to induce stress.~ In

addition, th~e subjects were free to terminate thie experiment

at any time, as stated in the written and verbal instructions

provided thiem (see Appendix C). However, all the subjects

road the entire passage; successfully completing the experi-

ment.


Experiment 11: Situational Stress


Situational stress can be defined as occurring when

an indijividual1 is exposed to a pa~rticular setting tha\t is

normally stressful. For example, a student taking an oral










oxamination ordinarily would be expected to experience d

form of situational stress. In this case, then, the stress

is caused by the particular act of taking the oral examina-

tion. Another example of situational stress is that per-

ceived by a dental patient awaiting treatment. Under these

cited conditions, individuals are threatened by the situa-

tion; it is thereby that stress is induced. In the case of

the student, the threat may be the possibility of not passing

the examination and having to confront thle consequences of

failure. Fear or anxiety resulting from thle perceived threat

of potential pain may be the cause of the stress in the dental

patient. In both cases, the circumstances create a threat

that results in the situation being stressful t~o Thr inidivi-

dual.



Population


Subjects for this experiment were chosen from students

enrolled in a public speaking course taught in ~the Speech

Department of the University of F~lorida. All students in

thle class were recorded; however, only the 17 withi MAACLL

scores of 13 or above were included in th~e analysis; nione

exhibited any speech/voice abnormalities. Su'ojects ranged

in age from 19 to 26 years with a mean age of 20.7 years;

there were 10 males and 7 females.










Speech Ma trials


Tw~o types of speech materials were obtained for this

experiment. The stress material consisted of recordings of

the actual speeches delivered by the students to an audience

of follow students in the public speaking class. The normal

sample consisted of ar reading of a modernization of R. L.

Stevenson's "An Apology for Idlers" obtained from only those

students with MAACL scores exceeding 12 for the stress con-

dition. The normal samples were recorded at IASCP in an IAC

sound thread oom



Experimental Procedure


The specially designed headset that would be used to

record. the speeches was demonstrated to th~e students

several days prior to th~e actual recording date. This

demonstration was done in order to explain the use of the

device and reduce any special anxiety that might be caused

by using the headset; however, thle students were not told

the actual purpose of the expe~rimnent. The students wore

thec headne:;t on t~wo occas:ions;-- ir.rst- while dlciveri~ng a

brief introductory speech. The pur-pos~e of- this procedure

was to accustom them to wearing the hieadlset and to further










reduce any anxiety they might feel which was alssociated

solely with using the headset. Therefore, it can be argued

that the stress experienced by the students during the re-

cording process would be the result of the public speaking

situation rather than the experimental procedure. Of

course, the students also wore the headset while their

speeches were recorded for the experiment.


Statistical Anailysis


The statistical technique used in the data analysis

was a matched-pairs t test.This procedure was selected be-

cause thie normal and stress speaking conditions were not

inicpendenit. In this instance, the two sample populations

consisted of the same individuals.

Since the samples were deliberately matched, if each

ea r f measures is treated as a single case, statistical

tests can be made legitimately. Instead of making a

difference-of-means-test, a direct pair-by-pair comparison

can be made by obtaining a difference score--in this case,

strons3 minus normal. Therofore, the null hypothesis is that:

there is no difference between the stress and normal speak-

ing conditions and it can be hypothesized that the mean of

the pair-by-pair difference in the population Pd is zero.








The problem then reduces to a singlo-samplee test of~ the

hypothesis that )ld = 0.

The difference between the normal and stress speaking

conditions was calculated for each of the measures obtained.

The mean xcd and standard deviat~ion sd were calculated for the

distribution of the differences, where

xdi
x = and
dN


S(a 2Xd
sd
Nd


The t test statistic is then calculated as






with N 1 degrees of freedom (df). Since the nu!ll

hypothesis was Ed = 0, the.test statistic reduces to

xd
td


A significance (alphul) level of 0.05 waI~s chosen for

th~e Last r:tatis;tic dlue to theu exploratlory na!ture oE~ thisr

study. In aIddition, a two-tailed test was used in order

to test for the direction of the shift in the parameter









under analysis. If the calculated value of' td is greater

than tc, where t, is the critical value of t at the 0.05

level with df = N 1, the null hypothesis rd = O is re-

jected. Furthermore, noting the sign of td (+) it can be

determined. whether the parameter increased (+) or decreased

(-) for the stress condition compared to the normal. For

a detailed discussion of matched-pairs t-tosts see Blalock

(1972)














CHAiPTER III


RESULTS OF LABORATORY STRESS EXPERIMENTS


This first study was conducted in order to compare

speech produced under stress to normal (nonstress) stress in

a highly controlled laboratory environment. The thrust of the

research is focused on an acoustical/temporal analysis of

speech produced under the two specified conditions. Specifi-

cally, the acoustical analyses carried out consisted of

various speech intensity (SI) and speaking fundamental fre-

quenciiy (Si;PF meast~uremenths--inclu dingl tili riaxim~um, modie, mean

and distribution of SI and the mean and distribution of SFF.

The measures within the timne domain consisted. of the mean

and standard deviation of speech rato (SR), speech/pause

ratio (SPR), speech time/total time ratio (ST/TT), the time-

energy distribution (TED), and determination of the number of

speech bursts and pauses. In addition, the number of dis-

fluencies occurring in the stress and nonstress speech sam-

ples also was calculated.

A matched-pairs t test was used in the statistical

analysis of the obtained data. F~or the matched-pairs t

test, the mean (RS) and standard deviation (Sd~) of the
55









differences between the normal and stress speaking condi-

tions--for each subject--are used in computing the test

statistic Ld. An alpha signiificanice level of 0.05 was used!

for all 1 cests. That is, if the significance level of tg

is 0.05 or less, the test scatiatic is considered signifi-

cant for thle purposes of thiis experiment. T:;e 0.05 alpha

level was chosen because the present study is exploratory in

nature and. seeks to determine which speech pa:rameter(s)

might be the vocal cue(s) to stress.


Speech Intensity (SI)


The first intensity characteristic analyzed as a possible

voca.l inclicatonr of StrP as was thp relatives mean1 iTntePnsity

(M1I) level. The results of the MI analyses are shown in

Table 1-A. As can be seen, thie overall MI was slightly

greater for the stress condiition than for the normal condi-

tion. That is, the mean differecrie between the speaking con-

ditions is 0.5 dB with a standard deviation of 0.8 dB. The

calculated c score for the difference between speaking con-

ditions wais found to be 2.07 (df = 11) which is nor signifi-

Canit ,L tLco O.05 level. ALC:hough: nona:;ignlificant, thie

positive Ld value probably indicates a trend for MI to in-

crease for speech produced under stress. Similar results








Table 1. Results of the Speech Intensity (SI) Ana~lyses
for the Laiboratory Stress Exsperimlent; All Values
are Expressed in dB Re: Imy. Values in Paren-
theses are the Standard Deviation (SD) and
Degrees of Freedom (df) for the t Scores.

Speaking Condition
Normal Stress Mean Dif-
(SD) (SD) ference(SD) td(df)

A. Mean Intensity (MI)
Males (N=7) 67.3 (1.7) 68.1(2.4) 0.8(0.7) 2.80(6)*
Females (N=5) 65.9(3.3) 65.9(3.1) 0.0(0.8) 0.00(4)
Overall (N=12) 66.7 (2.5) 67.2(2.8) 0.5(0.8) 2.07(11)

B. Maximum Intensity (MAXI)
Males (N=7) 79.1(3.6) 79.6(3.3) 0.4(0.5) 1.96(6)
Females (N=5) 78.4(2.9) 78.8(2.9) 0.4(0.5) 1.60 (4)
Overall (N=12) 78.8(3.2) 79.3(3.0) 0.4(0.5) 2.65(11)*

C. Mode of Distribution (MODIC)
Males (N=7) 73.3(3.8) 76.0C3.6) 2.7(3.0) 2.20(6)
Females (N=5) 71.8(6.5) 72.0(6.5) 0.2(1.1) 0.36(4)
Overall (N=12) 72.7(4.9) 74.3(5.2) 1.7(2.6) 2.17(11)


*Significant at thle 0.05 level.

t.05 = 2.447, df = 6

t05-276 df == 4


t.05 = 2.201, df = 11









were found for the males alone, the difference of 0.8J dB

was found to be significant (td = 2.80, dii = 6). Con-

versely, analysis of the female subjects showed no difference

between the normal anid stress speaking conditions. It

would appear that under the present experimental conditions

MII was not a robust indicator of stress.

The resullts of thle analyses of thle maxiimum intensity

(bYYC1) level are shown in Table 1-B. Analysis of the

overall results indicates a statistically significant in-

crease (0.4 dB) in M1AXI for the stress condition when it

was compared to the normal speaking condition. The maxi-

mum intensity produced under the stress condition was

79.3 dB, compared to 78.8 dU for the normal condition.

The difference in MAXI between the two conditions was found

to be n~onsigjnificant when controlled for thle subjects'

sex. However, the positive td values indicate at least

a slight tendency for MAXI to be greater for speaking

under stress than for normal speech.

Thie mode of the intensity lisLri~bution (MODI) also w~as

analyzed. The results of these anai;lyses may be found inl

Table 1-C. The 1.7 dB mean difference between the speaking










conditions was not found to be significant. Moreover, the

difference values were found to be 2.7 dB for the males and

0.2 dB for the females, neither of which were significant

at the 0.05 level. In each of the analyses MODI was negli-

gibly greater for the stressed speech than for normal speech.

Therefore, the tendency--although slight--is for MODI to

increase in stressed speech.

Finally, the entire intensity distribution (ID) was

analyzed for differences between speech produced under stress

and normally. The results of these analyses are shown

graphically in Figures 3-5.* The distributions were first

analyzed by sex, then the data were combined in order to

obtain the overall distribution. However, no statistically

significant differences were found. Indeed, the general

shape of the two distributions (normal and stress) remained

fairly constant across conditions. Therefore, analyzing the

entire intensity distribution did not discriminate between

normal speech and speech produced under stress.

In fact, MAXI was found to b-e the~c m~ost. powerf1-ul stress

indicator. The M9XI for the overr~l~l dalta was; significantly

greater for the stressed speech s;amples relative to normal

speech. The only other significant~ result; was found in MI

*The taibul~ated values for the distributions may be
found, in Appendix D.
















------NORMAL
-----STRESS



Xd= 0.01
Sd = 0.40
td = 0.14
af = 33

(t.05 = 2.035)


W
1)
Z



O

3
Z
W
v

W
a


I


50 55 60 65 70 75 80 85




INTENSITY (dB re:1 M V)

Figure 3. InteFnsity Distribution for the Miale Subjects
in the Laboratory Strese Experiment.





61








-----STRES~

Xd = 0.00
Sd = 0.22
t =0.00
d = 33
(t0 = 2.035)





o / \












2 L

I II I 1
[. 55- 60 6 7 5 O

INEST dBr: V
Fiur 4. ItniyDsrato o h eaeSb
jat nteMhrtrySrs xeiet


L
S


























































5


~ NORMAL
-----STRESS


Xd = 0.01
Sd. = 0. 24
td = 0. 24
d = 33
(t.05 = 2.035)


I I 1 I









50 55 60 65 70 75 80 8!



INTENSITY (dB re: 1MV)

Figu~rc 4. Overall Initensitl Distribuition; for All
Subjects in the Laboratoryy Stress Experi-
ment.










for thc madles. T'herefore, those intensity mncrsures were

not extremely indicative-of stre-ss--at least for this ex,-

periment.



SIpeaking Fundamental Frequency (SFF)


The other major acoustical parameter analyzed was

speaking fundamental frequency (SFF). SFF is ; measure of

the rate of opening and closing of the vocal folds. The

mean values of the obtained SFF can be seen in Table 2. The

overall mean SFF increased from 166.7 Hiz or 39.4 semitones

(ST) for the normal speaking condition to 39.7 ST or 169.3

Hz for stress. However, the mean difference between the

speaking conditions was not found to be significant. WJheni

analyzed by sex, the male subjects SFF increased from 35.4

ST (127.4 Hzt) for normal speech to 36.0 ST (132.0 H1z) for

stress, with a mean difference of 0T.6 ST (4.6 IHz). The

difference for the male subjects was found to be nonsignifi-

cant at the 0.05 level. Therefore, SFF for males appears to

increased :;l~ghtly for the nC1ross c~ondiL~i on while the femailes

exhibited no change at all in SFFI. However, there was a

slight tendency for SFF to be higoce For speech produced

under stress, although these increases we~re not; statistically

significant.












Table 2. Res-lts of the Speaking Fundamental' Frequency (SFF) Analyses for the Laboratory Stress
Experimecnt; values Are Expre~ssed inl Both Semit~ones (ST) anid Hocrtz (fGz); valur-s in
Pare.7-heses Are the Standard Deviaiionis (SD) and the Degrees of Proedom~ (df) for the
t Scrres.

;peaking conditions
Mea.n
Normal Stress Diff~ernce td
ST(SD) z H(SD) ST (SEH H z (SD) ST (SD) Hz (SD) S T(df~) Hzt(df)


Males (Ni = 7)


35.4(2.3) 127.4(16.4) 35.0(2,8) 132.0(20.R) 0,6(0,8) 4.6(6,6) 1.84(6)


1.71C(6)


Fcmales (N = 5) 45.C(2.3) 221.7(30.0) 45.0(2.3) 221.6(30.0) 0.0(0.4)-0.1(4.7) 0.00(4)

Overall(N, = 12) 39.4(5.41) 166.7 (53.2) 39.7 (5.3) 169.3 (51.9) 0.3 (0.7) 2.7(6.2) 1.42(11.)









In addition to the meain SFF, thec SFF distribution also

was analyzed. The resulting relationships may be found in

Figures6-8.* The SFF distributions for the two speaking

conditions were found to be very similar. The matched-

pairs t test for the distributions substantiated thiis

similarity. Th-e distributions were first analyzed by sex,

then the data were combined for the overall analysis.

However, differences between the normal and stress speaking

conditions were not found to be statistically significant.

Therefore, the SEF distributions were not very discrim~inating

between normal and stress speechi.


Sg"ch Rate (SR)


Speechi rate (SIR) wias defined as the number of syllables

produced per second. For the SR analysis the two minute

speech samples were divided into eight 15-second subsamples

and alternate subsamples utilized for the computation of

the mean and standard deviation of SR. The results of the

mean SR anallys~is may be found in Tablle 3. scnbsen

the mean value of SR~ increased from 4.41 syllablers/sec



Also see Appendix E for the tabsulated vle o h
SFF distributions.








--- NORMAL
--------STRESS


XcI = 0.01
Sil = 0.99
td ;= 0.06

(0.0 =2.042)


LLL
oII II l 1

So i 20 2 o 35 4 5
HZ 39 52 6 2 13 6 2

FUDAENA F QUENC
Fi~ ~ ~~~~~~~~~L ur .Sekn udmna rqec SF iti

buio for th aeSb e\ inteLb rt\

St es Ex eiment.









Xd = 0.01
Sd = 0.45
td = 0.12
df = 29
(t 2.045j)
.05


N ORMAL
----- STRESS


5


4






2






ST 20
HZ 52


25 30 35 40 45
69 92 123 165 220


5O 55
294 392


FREQUENCY


Figure 7. Speaking Fundamental Freque~ncy (SFF') Distri-
bultion for t~he Fama~~lee Sub~jects ini thle
Laboratory Stress ExpEriment.l


FUNDAMENTAL









-----NORMAL
----- STRESS


Xd = 0.01
Sd = 0.56
td, = 0.10
df =30
(t.05 = 2.042)


ST zo
Hz 52


25 Jo as 40 45 5o 5s
69 92 123 165 220 294 392


FUNDAMENTAL FREQUENCY

Overall Speaking Fuindamental Frequency (SFF)
Distribution for All SubjecLC in the Labora-
tory Stress Experiment,


Figure 8.






69









Table 3. Results of the Speech Rate (SIR) Analysis for the
Laboratory Stress Experimnent; SR Vallues are Ex-
pressed in Syllables Per Second; Values in Paren-
theses are the Standard Deviations (SD) and the
Degrees of Freedom (df) for the t Scores (N = 12).

Speaking Condition
Mean D~if-
Normal(SD) Stress(SD) ference(SID) dd)

Mean 4.41(0.24) 4.42(0.29) 0.01(0.20) 0.17(11)

Standard Deviation 0.25(0.15) 0.38(0.17) 0.13(0.21) 2.05(11)







for tboe normal speech snnmples to 41.42 syllablhess/se

for stress. H-owever, this increase was not found to be

statistically significant.

Similarly, the standard deviation of' SR also was

analyzed. The results may also be found in Table 3. The

SR variability was found to increase for speech produced

under stress although not significantly.

Although there was a slight tendency for b~oth the SR

mean and standard deviation to increase for the stressed

speech samples, the increase was niot sufficient to differ-

ontiate the normal and stress speech. Therefore, neither

of the SR measures were very utilitarian in indicating the

presence of stress.


Speech/Pause Ratio (SPR)


The speech~/pause ratio (SPR) was the next temporal

parameter analyzed. SPR is defined as th~e ratio of the per

cent of speech time to the per cent of pause time. The same

subsample segmen~tation used for SR was also utilized for

thie SPR analysis. That is, alterna~te 15 rsoc sub:;amples

wotro usdct for the comp~t~aL~ion of tLho :;PR mea!;n atnd standatlrd

deviation. The results of these analyses may be found in

Tib~le 4. Thle anlalysis of thle mean SPR resulted in a non-

significant difference between speech produced under stress

and normally.






71 .








Table 4. Results of the Speech/Pause Ratio (SPR) Analysis
for th? Laboratory Stress Experiment; Values in
Parentheses are the Standard Deviations (SD) and
the Degrees of Freedom (df) for the t Scores
(N = 12)

Speaking Condition
Mean Dif-
Normal(SD) Stress(SD) ference(SD) tddf

Mean 2.74(0.44) 2.90(0.46) 0.16(0.48) 1.11(11)

Standard Deviation 0.47(0.18) 0.58(0.24) 0.11(0.20) 1.82(11)










The results of the SPR standard devialtion analysis are

also shown in Table 4. Similar to the mean, the standard

deviation of SPR increased for the stress speech although

not significantly.

Therefore, SPR did not prove to be an adequate stress

indicator. However, both the mean and standard deviation

exhibited a slight tendency to increase in speech produced

under stress.


Time-Energy Distribution (TEDI



The time-energy distribution (TED) indicates the amount

of time a speech signal remains at a particular energy level.

An earlier investigation by Johnson (1978) indicated that

the tenth (or highest) energy level does not contribute to

the discrimination between waveforms. Therefore, in this

study only the lower nine energy levels were utilized in the

analysis. The TED values were obtained by segmenting the

two minute sample into eight 15-second subsamples and com-

puting the meain TXD value for very other segment. A

summary oE the resullts may be found in Figu~res 9-11.*

Inlitiallly, the analyses woro condiuctedl by sex in order

to examine possible differences due to sex; then the data


*See Appendix F for tabulated values of TED.










-- -STRESS


xg = 1. 37
90 Sd = 0.92
td = 4.22

(t.05 = 2.306)
BO




70 \





S60




so -


40 \




20


;r IO -


I tw11
2
ENRY EE
Fiur 4. Tm-nryDsrbtin(E)frteMl
Sujcsi h aoaoySrs xeiet






74

-- NORMAL
---STRESS


d =1.32
Sd =0.69
tcl = 5.42
dfr = 8
(t.05 = 2.306)


3 4 5 6 7 B
ENERGY LEVEL

TimY-Energy D~istribution (TIED) f'or the
Female Subjec~ts in the La~bora~tory Stress
Experiment.


Figure 10.








100 t- ----- NORMAL



Xd = 1.38I
90 - Sd = 0.76
tdl = 5.14
df = 8

(t.05 = 2.306)
80




70





60 \




50














if Illil
23 45678
ENRG LEVEL
Figur 11 vrl Tm-nry itiuio TD o
Al ujcsi h aoaoySrs x


per0mnt









were combined to obtain thec overall results. As can be seen,

the t scores were all1 positive and significant at the 0.05

alphla level. The t scores for the males is 4.22, 5.42 for

the females and 5.14 for the overall data.

The indication is that under stress a speaker will

maintain a higher intensity level for a greater portion of

his speaking time. This is illustrated in Figjures 9Y-11 by

the fact that the stress distribution is elevated relative

to the niormal distribution.



Speech Time/Total Time Ratio (ST/TT)


In addition to SPR, the ratio of the speech time to the

total samPle timne (ST,/TT) also was ~ianalyzed. The ST~/TT

measure is the percentage of the total speech sample

occupied. by speech. Thle results of the ST1/TT analyses are

shown in Table 5. The analyses included the mean and

standard deviation of ST/TT. The mean ST/TT was found to

increase, although not significantly. The mean difference In

ST/TT between the normal. and stress speaking conditions was

only 0.01. Similar results we~re found for the standard

dieviaitioni (or variability) of CT/Trr; that is;, the small

increase in variability for stressed speech was not signifi-

cant.






71







Table 5. Results of the Speech Time/Total Time (ST/TT) Ratio
Analysis for the Lab~oratory Stress Experiment;
Values in Parentheses are the Standard Deviations
(SD) and the Degree of Freedom (df) for the t
Scores (N = 12) .

Speaking Condition
M~ean Dif-
norma1(SD) Stress (SD) forence (SD) td(df)

Mean 0.73(0.03) 0.75(0.03) 0.01(0.04) 0.83(11)

Standard Deviation 0.03(0.02) 0.04(0.02) 0.01(0.02) 1.66(11)









In short, ST/TT was not found to be a very powerful

indicator of stress in this experiment. However, there was

a Londency for both the meani and standard deviation of ST/TT

to increase slightly for speech produced unider stress.



Speech Bursts and Pauses


Two other temporal measures obtained were the number

of speech bursts antd pauses. The results of these analyses

may be found in Table 6. As can be seen, the number of

speech bursts as well as the number of pauses was greater

for the stress speech samples. However, the increase was,

inconsequential (0.1) for both measures and not statistically

significant.






The final measure analyzed in this experiment was that

of disfluency. This measure was obtained utilizing the

total number of disfluences in the two minirute speech samples.

The mean values for the two speaking conditions are shown

inj Tab.10 7.( Ascn enen, thec 311num)r of dli:;f~luenciou~ w;Is

greater for thle stressed speech~ (7.7) thanir for thle normal

(5.2). In1 addition, the 2.5 increase was found tio be statis-

tically significant. Therefore, the affect of stress on














Table 6. Results of the Spceech Dunrsts aInd Paiuse Analyses
for thie LaboratorY Stretss Expecimnent; Value inl
Parenthesec is the Degrees of Freed~om (df~) for
the t Scores (NJ = 12) .

Speaking Condition
Mean
Normal Stress Difference td(df)

A. Speech B~ursts
Mean 40.0 40.2 0.1 0.10(11)
Standard Deviat~ion 5.7 .63.2

B. Pauses
Mean 40.6 40.7 0.1 0.10(11)
Standard Deviation 5.7 6.5 3.4












Table 7. Analysis Results of the Number of Disfluencies
for the Lab'oratory Stress Experiment; Value in
Parenltheses is the Degrees of Freedomn (df) for
the t Score (N = 12) .

Speakiing Condition
Mean
Normal Stress Difference td Mn

Mean 5.2 7.7 2.5 2.86 (ll)

Standard Deviation 3.1 4.6 2.9


*Significant at the 0.05 level.




Full Text

PAGE 1

AN ACOUSTICAL/TEMPORAL ANALYSIS OF EMOTIONAL STRESS IN SPEECH By JAMES WOODROW HICKS, JR. A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1979

PAGE 2

Copyright 1979 by James Woodrow Hicks, Jr

PAGE 3

This dissertation is dedicated to the memory of my parents James and Evelina Hicks

PAGE 4

ACKNOWLEDGME NTS I would like to express special thanks to Dr. Harry Hollien for his encouragement, guidance and help through some very trying times. He also counseled me during personal problems, which is deeply appreciated and "motivated" me when it was needed. Thanks are also extended to Drs . Howard Rothman, Arnold Paige, and W. Samuel Brown for taking the time to be members of my committee. Thanks are also due Dr. William Cutler of the Gainesville Veteran's Administration Hospital for the psychogalvanometer used to administer the electroshock in the laboratory experiment. Also, Dr. Tom Saine, Jim Harris, Ralph Partridge, Dr. Don Williams and Don Stacks, of the Speech Department, for allowing me to come into their classrooms to record their public speaking classes, thank you. I would like to acknowledge the entire staff of the Institute for Advanced Study of the Communication Processes (IASCP). None of us at IASCP would get much accomplished without their help.

PAGE 5

I would also like to thank all my friends for their encouragement during my work on this dissertation. Thanks also to all those who volunteered as subjects, specially for the electro-shock experiment, for my research. There are too few words to express my thanks to Cindy Dewey. She was always by my side and encouraged me in the worst of times. Our relationship did the most in helping me complete this dissertation.

PAGE 6

TABLE OF CONTENTS ACKNOWLEDGMENTS iv LIST OF TABLES x LIST OF FIGURES xv ABSTRACT xvii CHAPTER I INTRODUCTION 1 Psychology/Psychiatry 2 Forensic Communication 3 Diver Communication 6 Aerospace Coiuuiurij.cati.oris ....... / A Definition of Stress 10 Laboratory Induced Stress ..... 12 Situational Stress 13 Measures of Stress 14 Physiological Measures 14 Acoustical/Temporal Measures of Stress 17 Intensity 17 Fundamental Frequency (f ) 19 Temporal Measures 23 Disfluency 24 Subjective Judgements of Stress . . 25 Aural perception 25 Self-rating by subjects 26

PAGE 7

TABLE OF CONTENTS (Continued) Page Objectives 27 II METHOD 30 Acoustical Analyses 31 Speech Intensity (SI) 31 Speaking Fundamental Frequency (SFF). ....... 34 Temporal Analyses 36 Speech Rate (HR) .... 36 Time-Energy Distribution (TED) ... 37 Speech/Pause Ratio (SPR) 39 Speech Bursts and Pauses 39 Speech Time/Total Time (ST/TT) Ratio. 40 UiSiiUeuCy .............. 41 Subjects' Self-};.: Ling Stress ...... 42 Recording Procedure . 43 Experiment I: laboratory Stress . . . 46 Population. 47 • ' Speech Materials. ......... 47 Experimental Procedure 48 Experiment II: Situational Stress . . 49 Population. . , „ 50 Speech Materials, .51 Experimental ' rocuduru 51 Statistical Analysis 52 III RESULTS OF LABORA r J'(M:V STRESS EXPERIMENT . 55 Speech Intensity (SI) 56 Speaking Fundamental Frequency (SFF) . 63 vn

PAGE 8

TABLE OF CONTENTS (Continued) Pa£e Speech Rate (SR) 65 Speech/Pause Ratio (SPR) 70 Time-Energy Distribution (TED) 72 Speech Time/Total Time Ratio (ST/TT) . . 76 Speech Bursts and Pauses 78 Disfluency . . . 78 IV RESULTS OF SITUATIONAL STRESS EXPERIMENT . 82 Speech Intensity (SI) 83 Speaking Fundamental Frequency (SFF) . . 90 Speech Rate (SR) 92 Speech/Pause Ratio (SPR) ........ 97 Time-Energy Distribution (TED) . , , . . 98 Speech Time/Total Time Ratio (ST/TT) . . 104 Speech Bursts and Pauses . . . . 106 Disfluency 106 V DISCUSSION 110 Laboratory Stress Experiment .,_.... ' 110 Situational Stress Experiment. . „ , . . 114 Comparison to Previous Studies: SFF . . 118 Comparison to Previous Studies; Intensity , . . 120 Comparison to Previous Studies: Temporal 121 Conclusions. ......... „ , . , . 123 APPENDICES APPENDIX A--MULTIPLE AFFECT ADJECTIVE CHECK LIST 128 APPENDIX B — SUBJECT INFORMED CONSENT FORM ... 131 APPENDIX C — INSTRUCTIONS TO SUBJECTS 133 APPENDIX D — TABULATED VALUES FOR THE INTENSITY DISTRIBUTIONS. ...... 138 APPENDIX E — TABULATED VALUES FOR SFF DISTRIBUTIONS 145 APPENDIX F — TABULATED VALUES FOR" TIME -ENERGY DISTRIBUTIONS (TED) 152

PAGE 9

TABLE OF CONTENTS (Continued) Page REFERENCES 154 BIOGRAPHICAL SKETCH 159

PAGE 10

LIST OF TABLES Table Page 1 Results of the Speech Intensity (SI) Analyses for the Laboratory Stress Experiment; A. 1.1 Values are Expressed in dB Re: lmv. Values in Parentheses are the Standard Deviation (SD) and Degress of Freedom (df) for the t Scores 57 2 Results of the Speaking Fundamental Frequency (SFF) Analyses for the Laboratory Stress Experiment; Values are Expressed in Both Semitones (ST) and Hertz (Hz) ; Values in Parentheses are the Standard Deviations (SD) and the Degrees of Freedom (df) r-v: the t Scores 64 3 Results of rhe Speech Rate (SR) Analysis for the raboratory Stress Experiment; SR Values are Expressed in Syllables Per Second; Values in Parentheses are the Standard Deviations (SD) and the Degrees of Freedom (df) for the t Scores (N = 12) 69 4 Results of the Speech/Pause Ratio (SPR) Analysis for the Laboratory Stress Experiment; Values in Parentheses are the Standard Deviations (SD) and the Degrees of Freedom . (df) for the t Scores (N = 12) 7.1 5 Results of the Speech Time/Total Time (ST/TT) Ratio Analysis for the Laboratory Stress Experiment; Values in Parentheses' are the Standard Deviations (SD) and the Degree of Freedom (df) for the t Scores (N = 12). . . . 77

PAGE 11

LIST OF TABLES (Continued) T able Pag_e 6 Results of the Speech Bursts and Pause Analyses for the Laboratory Stress Experiment; Value in Parentheses is the Degrees of Freedom (df) for the t Scores (N = 12) 79 7 Analysis Results of the Number of Disfluencies for the Laboratory S tress Experiment; Value in Parentheses is the Degrees of Freedom (df) for the t Score (N 12) 80 8 Results of the Speech Intensity (SI) Analyses for the Situational Stress Experiment; All Values are Expressed in dB Re : lmv . Values in Parentheses are the Standard Deviations (SD) and the Degrees of Freedom (df) for the t Scores 84 9 Results of the Speaking Fundamental Frequency (SFF) Analyses for the Situational Stress Exper iment ; Vnic.es are Expressed in Both Semitones (ST) and Hertz (Hz) ; Values in Parentheses are the Standard Deviations (SD) an
PAGE 12

LIST OF TABLES (Continued) Table 12 13 14 15 16 17 D-l D-2 Results of the Speech Time/Total Time (ST/TT) Ratio Analysis for the Situational Stress Experiment; Values in Parentheses are the Standard Deviations (SD) and th3 Degrees of Freedom (df) for the t Scores (N = 17) ...... . Results of the Speech Bursts and pat Analyses for the Situational Stress Experiment; Value in Parentheses i f J the Degrees of Freedom (df) for the t Scores (N = 17) 105 107 Analysis Results of the Number of Disfluencies for the Situational Stress Experiment; Value in Parentheses is the Degrees of Freedom (df) for the t Score (T< -17) « . Summary of Results for the Laboratory n( vof^ fvnpT"inifint". D ased en Overall Mean Values for Each parameter .'...... Summary ofr.Results for the Situational Stress Experiment Based on Overall Mean Values for Each Parameter Comparison of the Overall Results for the laboratory and Situational Stress Experiments Based on Mean Values for Each Parameter Intensity Distribution Values for the rial'.-; Subjects in the Laboratory Stress Experiment; Intensity Values are Expressed in dB re : lmv . , .. In! (.Misity Distribution Values for the Female Subjects in the Laboratory Stress i:-rj.f' r i. merit ; Intensity Values are Expressed in dB re: lmv 109 111 115 119 13S 139

PAGE 13

LIST OF TABLES (Continued) Table Pa 3 e D-3 Intensity Distribution Values for All Subjects in the Laboratory Stress Experiment; intensity Values are Expressed in dB re : lmv ... 140 D-4 Intensity Distribution Values for the Male Subjects in the Situational Stress Experiment; Intensity Values are Expressed in dB re:lmv ^1 D-5 Intensity Distribution Values for the Female Subjects in the Situational Stress Experiment; Intensity Values are Expressed in dB re:lmv. l 42 D-6 Intensity Distribution Values for All Subjects in the Situational Stress Experiment; Intensity values are Expressed in dB rerlmv. l^5 £-1 Speakiny Fundamental Frequency y.3rF; Distribution for the JV^'le Subjects in the Laboratory Stress Experiment; Values of SFF are Expressed in Semitone (ST) Intervals 145 E-2 " Speaking Fundamental Frequency (SFF) Distribution for the Female Subjects in the Laboratory Stress Experiment; Values of SFF are Expressed i.n Semitone (ST) Intervals 14 ° E-3 Speaking Fundamental Frequency (SFF) Distribution for All Subjects in the Laboratory Stress Experiment; Values of SFF are Expressed in Semitone (ST) Intervals 147

PAGE 14

LIST OF TABLES (Continued) Table E-4 Speaking Fundamental Frequency (SFF) Distribution for the Male Subjects in the Situational Stress Experiment; Values of SFF are Expressed in Semitone (ST) Intervals . „ „ 148 E-5 Speaking Fundamental Frequency (SFF) Distribution for the Female Subjects in the Situational Stress Experiment; Values of SFF are Expre.'J'-.ed in Semitone (ST) Internals . .. „ 149 E-6 Speaking Fundamental Frequency (SFF) Distribution for All Subjects in the Situational Stress Experiment; Values of SFF are Expressed in Semitone (ST) Intervals .„,.... 150 F-l Mean Values !:'o«: 'r.lxc Time-Energy Distributions (TED) for the Laboratory Stress Experiment; Values are Luc Per Ccn^. Oj_ Time for Each of Nine Energy Levels . . 152 F-2 Mean Values i;or the Time-Energy Distributions (TED) for the Situational Stress Experiment; Va 'Lues are the Per Cent of Time for Each of Wine Energy Levels . . 153

PAGE 15

LIST OF FIGURES FM Wireless Microphone /Transmitter Used in Recording Process . . . „ 44 2 Specially Designed Headset Used to Hold FM Microphone/Transmitter (Show;;: in Place) in a Fixed Position Relative to Speaker ... 45 3 Intensity Distribution for t;\rMale Subjects in the Laboratory Stress Experiment 60 4 Intensity Distribution for the Female Subjects in the Laboratory Stress Experiment 61 5 Overall intensity Distribution for All Subjects in the Laboratory Stress Experiment . .. 62 6 Speaking Fundamental Frequency (SFF) Distribution for the Male Subjects in ilic Laboratory Stress Experiment 66 Speaking Fundamental Frequency (SFF) Distribution for the Female Subjects in the Laboratory Stress Experiment, ... 67 8 Overall Speaking Fundamental frequency (SFF) Distribution for All Subjects in the Laboratory Stress Experiment 68 C J Time-Energy Distribution (TED) tor; the Male Subjects in the Laboratory Stress Experiment , . 73 .1.0 Time-Energy Distribution (TED) for the Female Subjects in the Laboratory Stress Experiment 74 xv '.

PAGE 16

LIST OF FIGURES (Continued) Figure 11 12 13 14 15 16 17 18 19 20 Overall Time-Energy Distribution (TtlD) for All Subjects in the Laboratory Stress Experiment , . . Intensity Distribution for the Male Subjects in the Situational Stress Experiment , , . . Intensity Distribution for the Female Subjects in the Situational Stress Experiment , . . Overall Intensity Distribution fot; All Subjects in the Situational Stress Experiment . ........ Speaking Fundamental Frequency (SiVF) Distribution for the Male Subjects in the Situational Stress Experiment, „ . Snpakina Fundamental Frequence (S.FF) Distribution for the Female Subjects in the Situational Stress Experiment . Overall Speaking Fundamental Frequency (SFF) Distribution for All Subjects in the Situational Stress Experiment. . . Time -Energy Distribution (TED) lor the Male Subjects in the Situational Stress Experiment „ . . . Time-Energy Distribution (TED) f.r>r the Female Subjects in the Situation.."! Stress Experiment Overall Time-Energy Distribution (TED) for All Subjects in the Situational Stress Experiment Page 75 87 89 .93 94 .95 101 102 103

PAGE 17

Abstract of Dissertation Presented to the Graduate Council of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy AN ACOUSTICAL/TEMPORAL ANALYSIS OF EMOTIONAL STRESS IN SPEECH By James Woodrow Kicks, jr. December, 1979 Chairman: Harry Hoi lien Major Department: Speech Department The external manifestation of emotions has been of interest to researchers for many years. For example, as early as 1872 Charles Darwin discussed the use of facial and body movements as indicators of emotion in his book The . Exprcssio; of the Emotions in Man and Animals . However, Darwin's text is only descriptive in nature. Subsequently, various physiological measures have been correlated to emotional states; among these measures have been (1) heart rate, (2) galvanic skin response (GSR), (3) electroencephalogram (EEG) , (4) respiration and (5) blood pressure. However, more recently scientists have investigated speech parameters as possible indicators of the emotional state of the speaker. There are many instances in which knowledge about the emotional state of an individual would be desirable. It also would be advantageous to obtain this

PAGE 18

information without any direct, physical contact with the individual — as is currently required with the above mentioned physiological measures. A research thrust aimed at determining the acoustical/temporal speech parameters correlated to emotional stress should be useful to the area of speech communications. Most of the previous investigations have been limited in scope. For instance, many have analyzed only one parameter — primarily fundamental frequency (f ) — and/or had very small subject populations . The thrust of the current study focuses on an acoustical/temporal analysis of the effects of stress on speech. Included in the study were both laboratory induced stress and a "real" situational stress. The laboratory stress was induced via an electrical shock as the stressor. The situational stressenvironment consisted of a university level public speaking course in which the students were recorded while delivering speeches to an audience of their peers . Included in the analysis were several characteristics within the intensity, fundamental frequency and temporal domains. The following parameters were analyzed within the intensity domain: (1) maximum, (2) mean and (3) mode of the intensity distribution. The mean and distribution of

PAGE 19

speaking fundamental frequency (SFF) were analyzed within the frequency domain. The temporal parameters included: (1) speech/pause ratio, (2) speech rate, (3) the timeenergy distribution, (4) the number of speech bursts, (5) the number of pauses, and (6) speech time/total time ratio. In addition, the number of disfluencies in each speech sample also was measured. A comparison was made between speech samples produced normally and under stress. The comparison was based on the acoustical/temporal parameters analyzed. The results of the data analysis indicate that measurable acoustical/temporal changes do occur in speech produced ., , _ . _T L. .-— -T — ;_«?+ -V-Q O C rT~»QCS *"• V( Under Stress as COUlpci.L<3u i-u injj.iuct.i-, ihj.ii ~w.t~.~~ ~£ . However, the magnitude of these changes seems to be a function of the type of stress. For the situational stress experiment, 68.8% of the parameters differed significantly between the normal and stress speaking conditions, whereas only 18.8% changed significantly in the laboratory stressexperiment . It was found that for the laboratory stress paradigm all of the parameters increased for the stress speech, relative to normal. However, only the maximum intensity, time-energy distribution and the number of disfluencies

PAGE 20

increased significantly. That is, the stress induced by electro-shock did not alter the subjects' speech patterns to any great extent. However, significant differences were found to exist between the normal and stress speech samples in the situational stress experiment. Specifically, significant decreases were found for the intensity measures as well as the number of speech bursts and pauses. Conversely, the mean SFF and temporal parameters (except for speech rate) were found to increase significantly. Based on these findings, the general effects of stress on speech seem to: (1) decrease intensity, (2) increase SFF, (3) slightly decrease sn pf>ch rat-.e. and (4) decrease? the number of speech bursts and pauses, resulting in longer speech bursts. However, the variability in these parameters also increased indicating that the observed changes may not be uniformly consistent for all individuals. Therefore, baseline (normal) data maybe required for. an individual before the presence of. stress in . that individual can be detected.

PAGE 21

CHAPTER I INTRODUCTION For some time scientists have searched for parameters within the speech signal that might reveal the emotional state of the talker (Bonner, 1943; Cohen, 1961; Friedhoff et_al., 1964; Barland, 1973; Fairbanks & Hoaglin, 1941; Fairbanks & Provnost,. 1939; Hecker et al ., 1968; Kuroda et al., 1976; Silverman & Silverman, 1975; Simonov & Frolov, 1973, 1977; Williams & Stevens, 1969, 1972). A number of the cited experiments were conducted prj-or to current auvances in technology; hence, some of the parameters these scientists m ay have preferred to study were not amenable to their research. However, recent developments of equipment and techniques for the acoustical analysis of speech permit the investigation of issues previously not possible. It is well known that speech contains both linguistic and paralinguistic messages. That is, virtually any utterance conveys information on several levels— including that related to cultural background, regional dialect, country of origin, and possibly the general feeling about the

PAGE 22

2 statement being made by the speaker. Perhaps more important, a talker can transmit information via speech that reflects his current health and emotional state. There are many instances in which knowledge about the emotional state of an individual would be desirable. Further, there are situations where it would be advantageous to obtain this information without any direct, physical contact with the person being evaluated--f or example, in eierospace or diving operations. Indeed, research directed at determining the acoustical correlates of emotional stress should be useful to many of the sub-specialties within the speech communication area; among these specialties are Forensics, Acrcspacs ana Diver COiVimUiix cat ions , as well as FSyCuuiOyy/ Psychiatry. Psychology/Psychiatry Psychiatrists and psychologists appear to rely on the parallel encoding of linguistic and emotional information by their patients in order to treat these individuals (Friedhoff et al . , 1964; Ostwald, 1963). For example, Ostwald states, "whenever direct expression involves sound making — emotive sound making — acoustics has something tangible to contribute to psychotherapy" (p. -85). Thus, it would appear

PAGE 23

that Ostwald recognized the contribution to be made to psychotherapy by paralinguistic information encoded within the speech signal. A substantial benefit would accrue to psychotherapy if objective methods of evaluating the emotional state of the patient could be developed and used to complement subjective observations of the psychologist/ psychiatrist. In short, if the therapist could identify the emotion (fear, anger, or anxiety, say) being expressed by the patient, he should be able to provide more effective treatment. Forensic Communication Knovledae of emotional stress in speech is important to the area of Forensic Communication. For example, when a criminal is recorded making a bomb threat, it is desirable to determine his emotional state: Is it one where there is a high probability that he will follow through with the threat? Another illustration: It is obvious that when a suspect lies during interrogation, he is under stress. Can the acoustic speech signal be used to determine the likelihood that the suspect is lying in such cases? In an effort to answer these questions , "detectors " of voice stress recently have been developed. They are purported

PAGE 24

to be effective as "lie detectors"; they include the Psychological Stress Analyzer (PSA), Mark II, Hagoth (Bennett, 1977), Psychological Stress Evaluator (PSE) and Voice Stress Analyzer (VSA) (Kubis, 1973; McGlone, 1975; Almeida et al . , 1975). However, independent research has shown, for example, that at least two of these devices (the PSE and VSA) probably are not capable of the analysis claimed by their proponents (Barland, 1973; Kubis, 1973; McGlone, 1975). For example, the PSE reportedly evaluates stress by demodulating the inaudible, stress related. FM patterns known as "Muscle Micro Tremors (MMT) " (Almeida et al . , 1975). However, Almeida e t a 1 . conclude "our results, however, ao not confirm the theoretical basis ot the PSE" (p. 4). Speaker identification is another area within Forensic Communication that has been shown to be affected by stress (Hollien & Majewski, 1977; Doherty & Hollien, 1978). Hollien and Majewski found that the ability of the longterm speech spectra (LTS) technique to identify individuals was reduced somewhat under conditions of stress — induced by randomly applying electric shock to the subjects while they were speaking. When a fullband procedure (80-10,000 Hz) was used, the effect of stress was to decrease the identification scores by 8% over the unstressed (normal) speaking

PAGE 25

condition. Moreover, this difference (between the normal and stress identification scores) was greater (20%) when the samples were bandpassed (315-3150 Hz) to simulate a telephone transmission system. Therefore, it would appear that the presence of stress not only degrades the LTS speaker identification technique in a laboratory setting, but when "in-the-f ield" restrictions are imposed, this negative effect is magnified. Doherty and Hollien (1978) examined a set of three speaker identification vectors in the presence of speaker and system distortions; these vectors consisted of longterm speech spectra (LTS), speaking fundamental frequency (SFF) and speaking time (ST) . They used stress, induced by electric" shock, as one of the speaker distortions; system distortion was produced by subjecting the LTS vector to transmission distortion — specifically, to a limited passband of 315-3150 Hz. Doherty and Hollien conclude that, "while the described approach functioned adequately for the normal speaking condition, no vector (singly or in combination) adequately differentiated talkers when speech was distorted" (p. 1) . It is assumed that, while committing a crime, the criminal is under stress and his speech may be altered.

PAGE 26

Therefore, the changes that occur in the criminal's speech due to stress may influence the probability of making a correct identification. It would appear that knowledge about the effects of stress on speech is important to the area of Forensic Communication. Conversely, the development of, and attempts to use devices such as the PSE indicate the need for increased information about stress and its effect on speech. Diver Communication The relationship between stress and its effect on speech is also important to Diver Communication. Once the diver enters the water, he is in an alien environment that is dangerous and potentially fatal; thus, he is ipso facto in a stressful situation. Monitoring the diver's speech in order to determine his emotional state would add another safety feature to a diving operation. Moreover, knowledge of a diver's specific emotional state should be useful in anticipating and preventing diving accidents; in turn, a safer worhing environment should result. Safety measures are even more important in the case of saturation diving. In addition to the dangers and stresses of "shallow water" diving, the saturated diver remains

PAGE 27

7 submerged for long periods at great depths and is unable to surface safely in case of an emergency. Furthermore, a saturated diver tends to be assigned a heavier work schedule and lives in the highly restricted confines of an underwater habitat which leads to more stress — at least, when the situation is compared to conventional diving. Voice communication channels are usually available during most diving operations. The ability to use existing communications to monitor a diver's level of stress would add to the safety of diving operations without imposing additional burdens on the divers. Aerospace Communications The analysis of speech to determine emotional states _ also is important to Aerospace Communication. The emotional state of pilots and astronauts based on speech samples has been investigated by authors such as Williams and Stevens (1969), Simonov and -Frolov (1973) and Kuroda . et al. (1976). Williams and Stevens (1969) analyzed speech samples obtained during obviously (emotionally) stressful situations. Their recordings consisted of airto-ground communications between pilots and control tower operators during emergency (i.e., stressful) situations,

PAGE 28

plus a recording of the radio announcer describing the Hindenburg disaster. These authors obtained fundamental frequency (f ) contours of the stressed speech using narrowband sound spectrograms sampled at 0.15 sec intervals. They found differences in f Q contours of the speech samples recorded during the emergency situation as compared to samples before the emergency. The normal (nonstressed) sampleswere characterized by smooth, slow, and continuous changes in fundamental frequency as a function of time; the stress contours exhibited fluctuations that often were neither smooth nor continuous. Williams and Stevens state that when emotional stress is experienced, the speech of the individual may exhibit "sudden changes or jumps in fundamental frequency from one syllable to the next, and rapid up-and-down fluctuations may appear in the contour" (p. 1372) Furthermore, they found an increase in the mean f Q for the stressful condition as compared to normal speech — and the range of f was greater for stress. Williams and Stevens conclude that measurement of the median f and range of f may serve to classify whether an individual is undergoing emotional stress, providing, of course, his normal values are known.

PAGE 29

Simonov and Frolov (1973) estimated the emotional stress experienced by two cosmonauts during several phases of the flight of "Voskhod 2" and during training in heat and pressure chambers. They used a one-third octave spectral analysis within the range of the first formant (3001200 Hz) and compared these results to measures of heart rate (beats/minute) . These authors were able to differentiate the degree of emotional stress in their subjects about 85% of the time by means of this method. . Kuroda et al. (1976) analyzed the communications of pilots in 14 actual aircraft accidents, eight of them fatal, using a measure of fundamental frequency; specifically, the vibration space shift rate (VSSR) . VSSR is calculated by analyzing the spacing between the vertical striations on wideband sound spectrograms; specifically, the widest spacing between vertical striations during the normal phase of a flight is compared to the widest spacing encountered during an emergency situation. They found that the highest VSSR occurred during the initial phase of an emergency situation. Furthermore, when the emergency resulted in a fatality, Kuroda et al . found that a high VSSR had been maintained during the emergency. Therefore, Kuroda et al . (1976) conclude that "by analyzing the voice communications

PAGE 30

10 of a pilot involved in an emergency situation, v/e have a method of determining whether stress itself could have been a contributing factor in the outcome of that inflight emergency" (p. 533) . As can be seen from the above discussion, research on the manifestation of stress via speech is indeed applicable to numerous areas of speech communication. Its application ranges from assisting law enforcement agencies in their investigations to avoiding a possible fatality in diving and/or aerospace operations. However, the research conducted to date has been limited in its scope, leaving many questions unanswered. A Definition of Stress If research is to be carried out on stress, it would appear that the first step would be to operationally define this psychological state. Many attempts at such a definition have been reported. For example, in Psychological Stress: Issues in Research , Appley and Trumbull (1967) state, "It is further evident from the definitions cited that another area in which separation of psychological from physical aspects is required is that of threat" (p. 36) They also quote E. A. Haggard who stated, "An individual

PAGE 31

11 experiences: emotional stress when his overall adjustment is threatened, when his adaptive mechanisms are severely taxed and tend to collapse" (p. 39). Therefore, it would appear that threat is a major variable in determining emotional (psychological) stress — at least of certain types. That is, an individual must perceive a threat to his ego, integrity, values or goals. Lazarus (1966) states that the individual must "anticipate a confrontation with a harmful condition of s.ome sort" (p. 25) ; he goes on to say that "the strength of the stress response is determined by the strength of the intervening process, that is, by the degree of threat" (p. 26). Furthermore, emotional stress can be defined as a reaction to a stimulus condition; for example, Basowitz et al . (1955) concluded, "We should not consider stress as 'imposed' upon the organism, but as its 'response' to internal or external processes which reach those threshold levels that strain its physiological and psychological integrative capacities close to or beyond their limits" (pp. 288-289). Further, Appley and Trumbull (1967) suggest that stress "is a response state and that its induction depends on the mediation of some appraising, perceived, or interpreting mechanism" (p. 46). Therefore, it would appear that, in part, emotional stress is a mediated reaction to a threat to an individual .

PAGE 32

12 Often the terms stress and emotion are used interchangeable, but, are they synonymous? A given stressful situation may produce dissimilar emotions in different people as well as diverse emotions in the same individual at different times. Some emotions (grief, fear, anger) are invariably connected with stress, but that does not necessarily equate stress and emotion. "There are many emotions that have nothing to do with stress (love, joy, delight)" (Arnold, 1967, p. 50). Arnold has called the' emotions accompanying psychological stress "contending emotions"; they are primarily anger and fear and their combinations. Therefore-, as stated by Levitt (1967), stress "appears to be a kind of operator word which is applied in connection with emotion-evoking situations and reactions" (p. 15). Based on the above discussion, it would appear that a reasonable definition of stress is that it is a psychological state that is a response to a perceived threat and it will be accompanied by specific emotions . Laboratory Induced Stress Stress can be operationally defined in the laboratory; that is, as long as the definition meets the above criteria. Further, the definition can be based on, or related to, the

PAGE 33

13 use of a particular stressor of either physiological or psychological origin. For example, physiological stressors are usually employed to induce physical discomforts; they include electric shock (Farr & Seaver, 1975; Doherty & Hollien, 1978; Hollien & Majewski, 1977; Silverman & Silverman, 1975) as well as olfactory stressors (Ostwald, 1963; Farr & Seaver, 1975). On the other hand, procedures to induce psychological stress include having subjects (1) give a five-minute speech to a group of peers or (2) sit in a small room for ten' minutes with the object they are afraid of most (Farr & Seaver, 1975). Finally, it is important that the laboratory task be realistic compared to the task the subject will ultimately have to perform under real stress (as distinguished from laboratory stress) . in this case, the investigator's conclusions of subjects' behaviorunder real stress will be more valid. Situational Stress In this case, the experimenter lias not induced the stress; rather, the situation is normally stressful to the subjects. in situational stress, the consequences often are more severe than, for example, they would be if the subjects

PAGE 34

14 received a mild electrical shock. In this case, their lives might be threatened — as in the case of pilots or astronants in emergency situations (Kuroda et al . , 1976; Simonov & Frolov, 1973 and 1977; Williams & Stevens, 1969). In addition, the subjects may not have the option of avoiding the stressor. However, situational stress need not be life threatening; for example, students taking an oral examination or giving a public address might experience a high level of stress even though their life is not in danger. In this case the stress is of a psychological nature since the subjects are not being physically threatened. Therefore, the investigator can either create and induce stress in the laboratory or utilize the stress produced by a specific situation. In addition, he can use any of several measures to describe the effects of the stress. These measures can range from physiological to subjective/ perceptual . Measures of Stress Physiological Measures The external manifestation of emotions has been of interest to researchers for many years. For example, Charles

PAGE 35

Darwin (1872) discussed the use of facial and body movements as well as postures in his book The Expression of the Emotions in Man and Animals . However, Darwin's text is only descriptive in nature. Various methods subsequently have been used in an effort to measure the physiological correlates of emotional stress; these include (1) heart rate (Deane, 1961; Bowers, 1971. Bankart & Elliot, 1974); (2) galvanic skin response (GSR) (Bowers, 1971; Bankart & Elliot, 1974) ; (3) electroencephalogram (EEG) (Itil et al. , 1976); (4). respiration (Sovijarvi, 1974); and (5) blood pressure (Brod et al . , 1959; Blair et al. , .1959). While these measures have resulted in relatively accurate indicators of stress, they can introduce problems when used "in-the-f ield. " For example, physical contact with the subjects is required to obtain any of these measures. Such contact does not create serious problems in clinicalor laboratory applications but prolonged monitoring (of astronauts or divers, for example) can result in severe complications. Despite attempts to eliminate the problem, such as design improvements in the sensors and the use of creams or jellies' at the placement site, extended attachment of sensors may lead to skin irritations and other undesirable consequences. For this reason, it has been

PAGE 36

16 recommended that GSR, respiratory rate, EEG and other physiological measures be recorded only intermittently and the sensors removed after each recording. In addition, subjects' physical activity is restricted while the sensors are attached and he "becomes an observed subject who is by no means indifferent to the experiment in which he is involved" (Simonov & Frolov, 1977, • p. 23)— a condition that possibly can affect the resultant measures. Furthermore; attachment of electrodes or measuring equipment tends to distress a patient under psychiatric treatment. However, the therapist could easily tape record the therapy sessions as a standard procedure. The use of a tape recorder during therapy would disrupt the session minimally. The patient would simply accept the recording process as part of the psychiatric environment. Tape re' cording the therapy session would provide a permanent account of the session for later evaluation by the therapist. In addition, the recordings could be used as feedback for the patient, which can be important in psychiatric trea tment .

PAGE 37

17 Acous t ica 1/Tompora 1 Measures of Stress Analysis of the speech signal as an index of emotional stress is becoming a viable approach to research due to advances in the s tate-of-the-science of acoustic phonetic analysis equipment and techniques. However, the parameters of the speech signal, which are critical to the identification of stress, as yet have not been determined. Moreover, previous investigations have, focused primarily on the frequency domain of these signals (e.g.,fundamental frequency, range of fundamental frequency and formant structure) and have neglected intensity and temporal parameters. It would appear necessary that the parameters . of each of the.e detains be investigated as possible indicators of emotional stress. I ntensity Although it is a totally different type of stress, the study of research on linguistic* stress (emphasis) can be of aid to the Acoustic Phonetician. Linguistic stress has been found to influence the ' intensity of a stressed word. in a 1937 study by Tiffin and Steer, it was found that the average intensity of the stressed production of a word was 9.3 dB higher when compared to its unstressed

PAGE 38

18 production. Further, the stressed words were more intense than the unstressed about 71% of the time. On the syllabic level, Ortleb (1937) found that when compared to unstressed syllables, the emphasized syllables were more intense in both emotional and unemotional material; however, the intensity increase was more pronounced in the emotional material. Further, Schramm (1937) observed that the intensity of accented syllables was greater than the unaccented syllables in 78% of the words studied. Therefore, it would appear that intensity changes are important indicators of linguistic stress. When intensity features related to emotional stress have been-studied, the relationship between intensity and stress has not been found to be obvious. For example, when studying emotional stress, Hecker et al. (1968) did not find consistent intensity differences between their control (nonstress) and stress conditions. Six of their ten subjects exhibited small and inconsistent differences in intensity between the two conditions. Of the remaining four subjects, only one showed an increased intensity for the stress condition while the remaining three subjects _1 owe red their intensity level under stress. In a study by Friedhoff et al. (1964), subjects were instructed to respond no to a

PAGE 39

19 series of questions even though the correct response to some of the questions should have been yes; this technique was utilized to establish a mildly stressful lying situation. These authors report that average intensity increased under the condition of lying; however, variability was large, i.e., "some individuals give a consistently louder and some a consistently softer response when lying." Friedhoff et al aiso measured the intensity of the response two-six-nine under normal (nonstress) and stressed conditions, using electric shock as the stressor. They found that .maximum intensity shifted from the last word (nine) for the normal condition to the first word (two) for the stress condition. However, it is not .clear from 'their study whether they are reporting a trend across subjects or the results from only one subject. Therefore, it would seem that the manner in which intensity manifests emotional stress is not as well defined as it is for linguistic stress. However, intensity variability and the amount of intensity change might prove to be important indicators of emotional stress. Fundamental Frequency (f ) The f Q of speech also may be a parameter which is influenced by stress. For example, Fairbanks and Provonost

PAGE 40

2 (1939) studied the pitch characteristics of six actors, reading the same test passage and expressing five emotions-, contempt, anger, fear, grief, and indifference. They measured (1) pitch contours, 1 (2) pitch level, (3) pitch range, (4) rate of pitch change, (5) extent of pitch shifts, and (6) the number of changes in the direction of pitch movement per second. These authors report that they could distinguish among the five emotions they studied solely on the basis of the f Q measures. For example, they found that fear was identified as having the (1) highest median pitch level, (2) widest total pitch range, (3) largest excursion of all pitch shifts (together with anger) and of upward shifts within phrases, (4) largest change in pitch level between phrases, (5) fewest pauses during which no pitch change occurred, and (6) highest number of pitch changes per second of speaking time.Therefore, it would appear that fundamental frequency characteristics serve to differentiate among emotions. in an investigation of the affects of task-induced stress on speech, Ilecker et al. (1968) required ten subjects to read six meters and report the sum of the readings together with a test phrase. In this experiment they could Fairbanks and Provnost used the term pitch interchangeably with f Q -

PAGE 41

2 J vary the level of stress induced in the subjects by controlling the duration of the meter display. The results of this study indicated inconsistent changes in f Q across subjects. That is, some of the subjects raised their fundamental frequency while others lowered it as a function of increased stress. These authors point out that those subjects who lowered their fundamental frequency also spoke at a lower intensity level— as apparently these parameters were interrelated. Somewhat different results have been reported by Williams and Stevens (1969) who conducted their study using the air-to-ground communications between pilots and control tower operators during known stressful (i.e., emergency) situations. In an analysis o'f both pilots' and control tower operators' voices (for indications of stress), they obtained the level and range of f Q and f Q contours (from narrow band spectrograms) . They found that under increased levels of stress the pilots and air traffic controllers raised their f Q and showed "irregularities nd discontinuities" in f contours. Williams and Stevens (1969) also studied f usage in the voice of the announcer who described the Hindenburg zeppelin crash— both before and after this disaster. In addition to an increase in f Q a

PAGE 42

22 after the crash, they found that the announcer greatly reduced his inflectional pattern (variations in f Q ) during the period he was under high stress. In 1972 Williams and Stevens investigated the acoustical correlates of several emotions (anger, fear and sorrow) as portrayed by actors. In their analysis of these simulated emotions, they observed differences in mean f Q and range of f . They conclude that measurement of the median f Q and range of f Q for a speech sample serves to differentiate among emotions — assuming that thenormal f Q and range of f for the talker are known. On the basis of these experiments, it would appear that a speaKer's emotional state can be determined toy .analyzing the acoustic speech signal. However, it can be seen that disagreement remains as to the affect of stress on f as Hecker et al . (1968) report intersub ject. variability while Williams and Stevens (1972) indicate that f Q is raised when subjects undergo emotional stress. Some discrepancies also can be found in the data contrasting emotions based on f Q measures (Fairbanks & Provnost,. 1939; Williams & Stevens, 1972). Finally, when reporting changes in f Q , authors often describe: irregularities in f Q (rapid fluctuations) or irregular "bumps" in f

PAGE 43

23 contours, "irregular bumps in the contour, which might be interpreted as a kind of tremor" (Williams & Stevens, 1972) and "vibration of the vocal folds was too irregular" (Hecker et al. , 1968) . while the statements above are descriptive and observational in nature, analysis of these rapid fluctuations in f may prove useful in the detection of emotional stress. Thus, it would appear that the relationship between stress and its affect on speaking fundamental frequency requires further investigation. . Temporal Measures The temporal characteristics of speech have also been suggested as potential indicators of Doth emotional and linguistic stress. For example, increased duration of both words and syllables was found to be a major correlate of linguistic stress (Ortleb, 1937; Schramm, 1937; Tiffin & Steer, 1937) . Further, Fairbanks and Hoaglin (1941) found that simulated emotions (anger, fear, indifference, contempt and grief) could be differentiated to some extent on the basis of speaking rate, speech to pause-time ratio, and durations of speech and pauses. However, they report that the five emotions could be divided into two groups based on these temporal measures, with anger, fear,

PAGE 44

24 and indifference in one group and contempt and grief in the other. All three of the first group of emotions are characterized by rapid speaking rate and shortened durations of phonation and pauses, but did not differ significantly from each other. Contempt and grief, on the other hand, could be differentiated both from each other and from the other emotions. Although they were both .characterized by a slow rate, the rates (in simulating contempt) are produced by increasing the duration of both speech segments and pauses. On the other hand, the slow rate for grief results almost entirely from prolonged pauses, particularly those between phrases. Disf luency An additional parameter that may serve as an indicator of stress is speech fluency/disf luency . For example, Silverman and Silverman (1975) have defined this characteristic (disf luency) as follows: "repeating part of a word or an entire word, repeating a phrase, inserting extra sounds between words (ah, urn), changing the wording of a sentence, prolonging the sounds of a word, or breakingup the sounds of a word." In their study, Silverman and Silverman informed their subjects that an electric shock

PAGE 45

2 5 would be administered for each instance of disfluency during the reading of a 330 word passage. They found that under this stressful condition, normal speakers tended to become more fluent. That is, under stress the normal speakers were more accurate in their speech production as compared to the nonstress speaking condition. In short, it would appear that this feature may serve as an indicator of a speaker's emotional state. Subjective Judgements of Stress Stress can be quantified in other ways. These approaches include subjective measures such as aural/perceptual identification of stress by listeners and self-ratings by the subjects themselves of the level of stress they experienced. Aural perception It is generally agreed that the stress an individual experiences can often be detected merely by listening to his speech. Although aural/perceptual procedures are not as objective and analytical as the physiological, acoustical and temporal measures previously discussed, they remain

PAGE 46

26 valid dimensions for determining a speaker's emotional state. For example, Fairbanks and Provnost (1939) obtained correct aural perceptual identification of five emotions ranging from 66% to 88%. Therefore, they conclude that "emotions expressed by the voice alone are readily identifiable" (p. 104). Furthermore, Ilecker et al . (1968) used an aural/perceptual procedure in their study and found that listeners were able to correctly identify stressful responses from 66% to 83% of the time. An aural/perceptual evaluation is of major importance in any study using simulated emotions as speech material' in. order to verify the subjects' accuracy in affecting the emotions under investigation. Furthermore, the acoustical/ temporal parameters critical to the perception of stress could be determined by comparing speech samples judged to be normal and those produced under stress. Self-rating by subjects In addition to the measures previously cited (physiological, acoustical, temporal and aural/perceptual), selfratings by the subjects themselves are valuable in determining the level of stress they heive experienced (Acker & Reynolds, 1966; Zuckerman etal . , 1964). Indeed, this technique

PAGE 47

27 can be utilized to obtain an index of both levels of stress and possible intersubject differences in level. That is, what may be stressful to one subject may not stress another individual. Merely assuming that all the subjects had experienced a significant (or equal) level of stress could lead to erroneous conclusions based on the obtained data. For example, Farr and Seaver (1975) found that subjects rated various experimental protocols differently based on the level of stress produced. Therefore, selfrating by the subjects is -important in order to. insure that the subjects perceived a stress producing threat and to be able to quantify the level of perceived stress. Objectives The thrust of the current study is focused on an acous tical/temporal analysis of the effects of stress on speech. Included in the research protocol were both laboratory induced stress and nonlaboratory, or situational, stress. The laboratory stress was induced using a mild electric shock as the stressor. The situational stress condition consisted of speeches given by students enrolled in a university level public speaking course.

PAGE 48

28 It is possible that, with further research, the emotional and stress states of an individual could be determined by an analysis of speech features. With this objective in mind, the current study consisted of an analysis of several acoustical and temporal speech parameters produced under emotional stress. Included in the acoustical analyses are intensity and fundamental frequency. Furthermore, several characteristics of eeich acoustical parameter were analyzed. For example, the followiny characteristics within the intensity domain were measured: (1) maximum' intensity, (2) mean intensity, (3) the mode of the intensity distribution, and (4) the intensity distribution itself. Similarly, Seveicio. c iiar ac Ler is t ics wxl.iH.ii i_rie ceiuporax uoriiaxn axoc were analyzed. The temporal characteristics included; (1) speech/pause ratio, (2) speech rate, (3) the timeenergy distribution, (4). the number of speech bursts,(5) the number of pauses, and (6) speech time/total time ratio. In addition, the number of disfluencies in the two-minute speech sample also were measured. A comparison was made between normal speech and speech produced under stress. The comparison of these two speaking conditions was based on the acoustical/temporal parameters analyzed. The ultimate goal of the present study is to

PAGE 49

29 add to the corpus of knowledge concerning the effects of stress on speech. However, the specific goals of this research may be stated in the form of several questions. They are as follows: 1. Which of the many selected acoustical/temporal parameters,, found within the speech signal, might serve as indicators of emotional stress? 2. Will stressful speaking conditions significantly increase or decrease the levels of the measured parameters? 3. is the resultant change due to stress ina specific parameter consistent for all speakers?

PAGE 50

CHAPTER II METHOD The objective of this investigation is to compare speech produced under .stress to nonstressed (normal) speech. The analyses carried out included both acoustical and temporal measures as well as the level of disfluency observed and a self-rating (by the subjects) of the stress they experienced. The acoustical analysis consisted of several measures based on (1) speech intensity (SI) and (2) speaking fundamental frequency (SFF). The temporal measures analyzed were (1) speech/pause ratio (SPR) , (2) speech rate (SR) , (3) the time-energy distribution (TED), (4) the number of speech bursts, (5) the number of pauses, and (6) speech time/total time ratio (ST/TT) . In addition, the number of disfluencies was determined for both speaking conditions (stress and nonstress); two minute speech samples were used for all the analyses. The self-rating by the subjects obviously was based on each subject's overall reaction to the experimental procedure and was a criterion for subject inclusion. 30

PAGE 51

31 The analyses utilized were common to two experiments, i.e., stress induced by either (1) electroshock or (2) public speaking was examined. Therefore, the types of analysis carried out will be discussed first; these descriptions will be followed by reviews of the procedure used for each experiment. Acoustical Analyses Included among the acoustical analyses were several parameters within the domains of speech intensity and speaking fundamental frequency. They are as follows. S peech Intensity (SI) • First among the intensity characteristics to be analyzed was mean intensity (MI) . In addition to the mean, the maximum intensity (MAXI) also was analyzed. MAXI is defined as the highest intensity produced by each subject for the entire two minute speech sample. The mode of the intensity distribution (MODI) defined as the intensity level most frequently produced, also was determined for each speech sample. Additionally, the entire intensity distribution (ID) for each speech condition was analyzed for differences between the normal and stress speech samples.

PAGE 52

3 2 A computer aided procedure war. used to obtain the various intensity measures; a Digital Equipment Corporation (DEC) PDP-8i minicomputer and analog -digital (A/D) converter were used for this purpose. First, the envelope of the speech signal was obtained using a rectifier/integrator circuit. The speech envelope was negative dc shifted in order to make full use of the A/D range (-1.0 to 1.0 V). The computer software (ADCNV) , controlling the A/D converter, allowed the computer operator to specify the parameters of sampling rate and the length of the speech sample. The sampling rate used was 50 samples per second and the sample length was 120 sec (2 min) . This rate was determined by testing numerous drf ferent rates in oroer co obtain the lowest sampling rate that would accurately represent the speech envelope. The output of the A/D converter was then stored on magnetic tape (DECTAPE) and the analysis program (INTANA) was implemented. INTANA was written to give the operator substantial control over several program parameters. The first parameter the program requires the operator to enter is the A/D value of the dc shift (AD ZERO) used during the inputting of the data. Next, the operator specifies the A/D value of a dc calibration voltage (AD CAL) . This calibration value is used by INTANA to

PAGE 53

3 3 convert A/D units to dB values of intensity. The next parameter requested by IN TANA defines the calibration voltage as a multiple (MULT) of the dB reference level selected by the operator. This input was followed by a dB adjustment (DB ADJ) used to compensate for any amplification or attenuation of the original signal. That is, if the signal had been amplified at the time it was recorded, the operator would enter the negative value of the amplification level (in dB) . If, on the other hand, the signal had been attenuated, the value of DB ADJ' would be positive — in order to add back the amount of the original attenuation. Following the D3 ADJ, the program requested a dB cutoff (DB CUTOFF). Any signal level below DB CUTOFF was considered to be either noise or a pause and was not included in the computation of the various intensity measures. The intensity analysis was then begun using the A/D equivalent of DB CUTOFF as well as the 'previous operator specified parameters. The INTANA output included the A/D values of the minimum (AD MIN) and maximum (AD MAX) levels of the stored data, as well as the actual number of sample points for AD MIN and AD MAX and the corresponding percentage of the total data for each. The total number of valid (i.e., speech) and rejected (i.e., noise or pauses) data points and their corresponding

PAGE 54

3 4 percentages were indicated by the output of INTANA as were the maximum (DB MAX), range (RANGE), mean (DB MEAN) , and standard deviation (STD DEV) of intensity (in dB). Next, a distribution table of intensity levels was printed; it listed the dB value, the number of sample points at each level, and the corresponding percentages. The final output was a histogram of the intensity values. In summary, the intensity measures used are the mean (MI), maximum (MAXI) , the mode of the distribution (MODI), and the overall distribution (ID). As stated, these intensity measures were utilized in comparing speech produced under stress and nonstress conditions. The rationale for using so many intensity measures is that it remains to be determined which speech intensity measures are critical to detecting the presence of stress. Therefore, it was judged that as many intensity parameters as possible should be investigated and the critical measures identified. Speaking Fundamental Frequency (SFF) Speaking fundamental frequency (SFF) is a measure of the rate of opening and closing of the vocal folds. In the present study the SFF measures of interest, i.e., mean SFF

PAGE 55

3 5 and the SFF distribution, were compared as a function of the stress and normal speaking conditions. The spscified SFF measures were obtained using the Fundamental Frequency Indicator (FFI) available at the Institute for Advanced Study of the Communication Processes (IASCP) . FFI is a digital fundamental frequency tracking device consisting of low-pass filters with cutoffe at halfoctave intervals coupled with high-speed switching circuits. FFI produces a series of pulses, coinciding with the fundamental period of a complex speech waveform. These pulses are then processed by a DEC PDP-8i minicomputer using an internal clock to mark the time interval between pulses and this data is processed to yield the geomecricai mean frequency and standard deviation of the fundamental 'frequency distribution. In addition, FFI produces a SFF distribution table and a plot of the SFF distribution. The distribution table produced by FFI lists the following data : SFF in (1) semitone (ST), (2) absolute frequency (Hz), (3) the number of occurrences of each value of SFF, and (4) the per cent of the total phonation time each measure of SFF was produced. The value of SFF in Ilz and ST and the per cent of occurrence were the values used from the SFF distribution table.

PAGE 56

30 The mean SFF , both within and across sexes, was compared between speech samples produced under normal and stressed conditions. In addition, the SFF distribution for each subject was used to compute a composite SFF distribution for each sex. That is, a composite distribution was determined for the males and one for the females and each used to compare the two speaking conditions within sexes . Temporal Analyses In addition to the acoustical parameters discussed above, temporal aspects of speech also were investigated as possible indicators of emotional stress.The temporal characteristics analyzed consisted of (1) speech rate (SR) , (2) a time-energy distribution (TED), (3) speech/ pause ratio (SPR) , (4) the number of speech bursts, (5) the number of pauses, and (6) speech time/total time ratio (ST/TT) . Speech Rate (SR) Speech rate (SR) is defined as the number of syllables produced per second. This temporal feature was measured at the syllabic level due to varying word lengths; obviously,

PAGE 57

37 polysyllabic words will take longer to produce than monosyllabic words. Therefore, using word boundaries would result in misleading data. The syllabic level of measurement will become most important for the stress speech samples obtained in the public speaking experiment (to be discussed later) because the intersubject speech samples will be different; i.e., they are not the same samples. The obtained two minute samples were divided into 15 sec segments and SR calculated for each segment by dividing the number of syllables in each segment by 15 sec. Finally, the mean and standard deviation of SR for the two minute samples were calculated. Time-Energy Distribution (TED) The time-energy distribution (TED) parameter originally was developed in 1978 by Johnson for a speaker identification study. "In general terms,. TED reflects the total time a talker's speech intensity remains at a specific energy level (relative to his peak amplitude). It also provides indication of the speaker's speech pattern with respect to speech bursts and pause periods" (Johnson, 1978, p. 39). In the present study, TED was added to the other measurements in an attempt to discriminate between stress and normal speech.

PAGE 58

3G TED was obtained utilizing a PDP-Si minicomputer and A/D converter. initially, the envelope of the speech sample was generated using a rectifier/integrator circuit. The speech envelope then was digitized (A/D) and analyzed in real time using software developed by Johnson (1978) . This software segments the speech signal into ten equal intensity levels relative to the maximum intensity produced by the speaker. Based on this segmentation, the TED program computes; (1) thenumber of speech bursts, (2) speech bursts per sec, (3) time in sec, (4)per cent of the total time, (5) mean time in millisec, and (6) standard deviation in millisec of each speech burst for each of the ten intensity levels. .The mean pause periods and the number of pauses were the direct reciprocals of the speech bursts. In this . manner, TED determines the distribution of both the speech bursts as well as the pause periods. The speech burst distribution was used in order to analyze differences between speech samples produced under the two speaking conditions (stress and no stress) . In addition, the TED output also was used to determine the speech/pause ratio (SPR) , the number of speech bursts, the number of pauses, and the speech time/total time (ST/TT) ratio (see below).

PAGE 59

39 S peech/Pause Ratio (SPR) The speech/pause ratio (SPR) is the ratio of the per cent of speaking time to the per cent of pause time. A pause is defined as that periou of time during which an individual is not. producing any audible sound. SPR can be readily obtained using the TED printout; that is, SPR can be calculated by using the percent of speech time and pauses for intensity level one, the lowest level. Each two minute speech sample was divided into eight 15 sec segments and SPR was calculated for every other segment beginning with the first segment. This approach allowed the investigation of those SPR changes within each sample that may have been "averaged out" by examining the entire two minute sample. Furthermore, this segmentation permitted the calculation of the mean and standard deviation of SPR for the entire sample. Speech Bursts and Pauses in addition to SPR, specific speech bursts and pauses were counted in order to determine possible differences between normal (nons tress) speech and speech produced under stress. The rationale for this analysis was as follows: a

PAGE 60

40 reliance only on SPR would result in certain confusions. That is, one speech burst of 20 sec and four bursts of five sec each would both result in a SPR of 0.5 for a 40 sec sample, when in fact, the two samples are different. The number of speech bursts and pauses were obtained from the lowest TED intensity level. This intensity level distinguishes between portions of the total sample that include vocalizations and those that do not (i. e., pauses) . Speech Time /To Lai Tim e ( ST/TT) Ratio The speech time/total time ratio (ST/TT) is the ratio of the amount of time in which speech' occurred to the total time of the spaech sample. This measure compares the speech time of a given sample to the total time of that sample. The ST/TT ratio was calculated utilizing the TED output. Both the ST and TT of each speech sample were obtained from level one of the TF.D. The same segmentation of the total sample used to calculate SPR and SR applies to ST/TT. That is, each

PAGE 61

41 sample was divided into eight 15 sec sub-samples and starting with the first sub-sample, alternate sub-samples were used to calculate the mean and standard deviation of ST/TT. Disf luency The final parameter analyzed was speech disf luency. This parameter is neither an acoustical measure nor entirely a temporal measure. While disf luency does relate to the temporal domain, in the present study it was not measured as a function of time. Instead, each occurrence of a disfluency was marely counted. Therefore, this parameter will be discussed separately from the acoustical and temporal parameters. The definition of disfluency used in the present study is: "repeating part of a word or an entire word, repeating a phrase, inserting extra sounds between words (ah, um) , changing the wording of a sentence, prolonging the sounds of a word, or breaking-up the sounds of a word" (Silverman, & Silverman, 1975, p. 353). The present author and a speech pathologist listened to each two minute sample and

PAGE 62

A A counted each time the speaker was cl is fluent. Included among the types of disfluencies measured were "audible pauses"; that is, sounds inserted between words (e.g., ah and urn) . These audible pauses would have been measured as speech in the SPR measure discussed above. Inclusion of audible pauses in the SPR would have artificially increased the SPR. This possibility would have been obscured without separately examining disfluencies. In addition, although the experimental paradigm was different from the present study, in their study Silverman and Silverman found that disfluency and the threat of electric shock are correlated. Therefore, it would seem that the number of disfluencies may serve as an indicator of stress. Subjects' Self-Rating Stress The individuals in both experiments rated the amount of stress they experienced in order to select subjects for inclusion in each experiment. This re If -rating was accomplished using the Multiple Affect Adjective Check List (MAACL) (Zuckcrman c t a 1 . , 1964). The MAACL has three scales in order to measure: (1) anxiety, (2) hostility,

PAGE 63

43 and (3) depression. However, only the anxiety score was used for subject selection in the present study. The MAACL has a maximum anxiety score of 21, with normal (nons tressed) subject groups scoring 4-7 and individuals under stress averaging scores of about 15. Each subject completed the MAACL response form prior to the initiation of the speaking task under the stress condition. Only those individuals with scores of 13 or greater were included in the experimental population in order to insure that subjects had achieved a significant level of stress. A sample of the MAACL may be found in Appendix A. Recording Procedure The recording procedure will be discussed separately from the experimental descriptions since it was common to both experiments. Specifically, a FM wireless microphone/ transmitter was used to record the speech samples as can be seen in Figure 1. A special headset was designed and fabricated in order to position the microphone in constant relationship . to the speaker. This headset with the microphone in place can be seen in Figure 2. The rationale for controlling the relationship between the microphone and the subject was to eliminate intensity variations resulting from

PAGE 64

r O 44 CD = — a

PAGE 65

4 §»•?•* -lv-^ ••..V«.'i Figure 2. Specially Designed Headset Used to Hold I'M Microphone/Transmitter (Shown in Place) in a Fixed Position Relative to Speaker.

PAGE 66

46 head movement. The speech signal from the microphone/ transmitter was received on a FM tuner which was coupled to a Sony TC-353D tape deck via a Hewlett-Packard 350D variable attenuator. The output of the FM tuner and the input to the tape deck were maintained at fixed levels such that attenuation was required in order to not overdrive the input to the tape deck. Therefore, the recording level was controlled with the attenuator, which facilitated maintenance of a constant gain in the recording system for both the normal and stress speech conditions for a given subject. Maintaining a constant gain between speaking conditions permitted intensity comparisons between conditions. Additions .lly. this record inn technique v.'3R chosen in order to minimize the affect of the recording procedure on a subject's level of stress — particularly in the situation stress experiment. Furthermore, it was judged that. i.i. allowed the subjects freedom of movement not possible with conventional recording techniques while maintaining experimental control. Experiment I: Labo ratory Stress In a laboratory setting the experimenter lias greater control over the stressing environment as compared to "real" or situational stress. Furthermore, it would seem reasonable

PAGE 67

(VI to first Lost a new procedure under highly controlled conditions before advancing to nonlaboratory , in-the-field environments. Therefore, the first experiment in this study was conducted in a laboratory environment. Laboratory stress, as used in this experiment, is defined as a deliberately contrived stress induced in a laboratory environment. In this particular case an electric shock, was used as the stressor. Population The subjects utilized in this experiment were recruited from members of the Institute for Advanced Study of the Communication Processes (IASCP) and the Speech Department of the University of Florida. They ranged in age from 20 to 31 years, with a mean age of 23.9 years. The total population included 7 males and 5 females; none exhibited any speech/voice abnormalities. All subjects scored 13 or above' on the MAACL anxiety scale under the stress condition. Speech Ma tor ia].s Speech materials consisted of two readings of a modernization of R. L. Stevenson's "An Apology for Idlers." First the subjects read the passage with no electric shock

PAGE 68

48 administered (i.e., no stress induced). This reading constituted their normal speech sample. Next, the subjects randomly received an electric shock while reading the standard passage, and this was used as their stressed speech sample. "An Apology for Idlers" was selected as reading material based on the time required to read the passage ; approximately 3-3.5 minutes. A passage of this length was needed in order to extract the two minute speech sample used in this experiment (see Appendix C) . Experimental Procedure Recordings for this experiment were obtained by utilization of the procedures described above; subjects were recorded in an Industrial Acoustics Company (IAC) sound treated room. Only the FM receiver was placed in the IAC room with the subject, while the remainder of the recording equipment was operated outside the room and coupled to the receiver via the IAC room's patch panel. The subjects' first task was to read "An Apology for Idlers" with no stress being induced (i.e., no shock was administered) in order to obtain a normal speech sample. At this juncture, the electroshock procedure was explained and the electrodes were placed on the index and ring fingers of

PAGE 69

49 the subject's hand. Once the electrodes were in place, the subject was asked to sign a Subject Informed Consent Form (see Appendix B) and complete the MA ACL test. The subject was told that the number of shocks (one to seven) and when they would occur had been randomly assigned, but that at least one shock would be administered. In fact, all subjects received seven shocks given on the same preselected words during the reading. A Grason-Stadler Psychogalvanometer , Model 4, constant current shocking device was used to administer the electrical stressor. A current level of 2.5 ma was delivered manually. Of course, this procedure did not harm the subjects; it only caused enough discomfort and threat to induce stress. In addition, the subjects were free to terminate the experiment at any time, as stated in the written and verbal instructions provided them (see Appendix C). However, all the subjects read the entire passage; successfully completing the experiment . E xperiment II; Situational Stress Situational stress can be defined as occurring when an individual is exposed to a particular sotting that is normally stressful. For example, a student taking an oral

PAGE 70

examination ordinarily would bo expected to experience a form of situational stress. In this case, then, the stress is caused by the particular act of taking the oral examination. Another example of situational stress is that perceived by a dental patient awaiting treatment. Under these cited conditions, individuals are threatened by the situation; it is thereby that stress is induced. In the case of the student, the threat may be the possibility of not passing the examination and having to confront the consequences of failure. Fear or anxiety resulting from the perceived threat of potential pain may be the cause of the stress in the dental patient. In both cases, the circumstances create a threat that results in the situation being stressful to "the individual. Population Subjects for this experiment were chosen from students enrolled in a public speaking course taught in 'the Speech Department of the University of Florida. All students in the class were recorded; however, only the 17 with MAACL scores of 13 or above were included in the analysis; none exhibited any speech/voice abnormalities. Subjects ranged in age from 19 to 26 years with a mean age of 20.7 years there were 10 males and 7 females.

PAGE 71

Speech Materials Two types of speech materials were obtained for this experiment. The stress material consisted of recordings of the actual speeches delivered by the students to an audience of fellow students in the public speaking class. The normal sample consisted of a reading of a modernization of R. L. Stevenson's "An Apology for Idlers" obtained from only those students with MAACL scores exceeding 12 for the stress condition. The normal samples were recorded at IASCP in an IAC sound treated room. Experimental Procedure The specially designed headset that would be used to record the speeches was demonstrated to the students several days prior to the actual recording date. This demonstration was done in order to explain the use of the device and reduce any special anxiety that might be caused by using the headset; however, the students were not told the actual purpose of the experiment. The students wore the headset on two occas i ons — f i rs t , while delivering a brief introductory speech. The purpose of this procedure was to accustom them to wearing the headset and to farther

PAGE 72

52 reduce any anxiety they might feel which was associated solely with using the headset. Therefore, it can be argued that the stress experienced by the students during the recording process would be the result of the public speaking situation rather than the experimental procedure. Of course, the students also wore the headset v/hile their speeches were recorded for the experiment. Statistical Analysis The statistical technique used in the data analysis was a matched-pairs t test. This procedure was selected because the normal and stress speaking conditions were not independent. In this instance, the two sample populations consisted of the same individuals. Since the samples were deliberately matched, if each p air of measures is treated as a single case, statistical tests can be made legitimately. Instead of making a difference-of -means test , a direct pair-by-pair comparison can be made by obtaining a difference score— in this case, stress minus normal. Therefore, the null hypothesis is that there is no difference between the stress and normal speaking conditions and it can be hypothesized that the mean of the pair-by-pair difference in the population p. d is zero.

PAGE 73

The problem then reduces to a single-sample test of the hypothesis that p^ 0. The difference between the normal and stress speaking conditions was calculated for each of the measures obtained. The mean x d and standard deviation s d were calculated for the distribution of the differences, where h — . dx , x -, = x and \|? (x d x d» The t test statistic is then calculated as fc d = X d " ^ 'd f N 1 with N 1 degrees of freedom (df) . Since the null hypothesis was ^ d = 0, the. test statistic reduces to *d dy \| N 1 ' A significance (alpha) level of 0.05 was chosen for tiie test statistic due to the exploratory nature of this study. In addition, a two-tailed test was used in order to test for the direction of the shift in the parameter

PAGE 74

:54 under analysis. If the calculated value of t d is greater than t c , where t is the critical value of t at the 0.05 level with df = N 1, the null hypothesis /i d = is rejected. Furthermore, noting the sign of t d (t) it can be determined whether the parameter increased (+) or decreased (-) for the stress condition compared to the normal. For a detailed discussion of matched-pairs t-tests see Blalock (1972) .

PAGE 75

CHAPTER III RESULTS OF LABORATORY STRESS EXPERIMENTS This first study was conducted in order to compare speech produced under stress to normal (nonstress) stress in a highly controlled laboratory environment. The thrust of the research is focused on an acoustical/temporal analysis of speech produced under the two specified conditions. Specifically, the acoustical aneilyses carried out consisted of various speech intensity (SI) and speaking fundamental frequency (SFF) measureuieuLb — including the maximum, mode, mean and distribution of SI and the mean and distribution of SFF. The measures within the time domain consisted of the mean and standard deviation of speech rate (SR) , speech/pause ratio (SPR) , speech time/total time ratio (ST /TT) , the timeenergy distribution (TED), and determination of the number of speech bursts and pauses. in addition, the number of disfluencies occurring in the stress and nonstress speech samples also was calculated. A matched-pairs _t test was used in the statistical analysis of the obtained data. For the matched-pairs _t test, the mean (x^) and standard deviation (S d ) of the 55

PAGE 76

56 differences between the normal and stress speaking conditions—for each subject—are used in computing the test statistic tjjj. An alpha significance level of 0.05 was used for all _t tests. That is, if the significance level of t_j is 0.05 or less, the test statistic is considered significant for the purposes of this experiment. The 0.05 alpha level was chosen because the present study is exploratory in nature and seeks to determine which speech parameter (s) might be the vocal cue(s) to stress. Speech Intensity (SI) The first intensity characteristic analyzed as a possible vocal indicator of stress was th^> relative me-~vn i n hens it" (MI) level. The results of the MI analyses are shown in Table 1-A. As can be seen, the overall MI was slightly greater for the stress condition than for the normal condition. That is, the mean difference between the speaking conditions is 0.5 dB with a standard deviation of 0.8 dB . The calculated t. score for the difference between speaking conditions was found to be 2.07 (df = 11) which is not significant lit the 0.0'j level. Although nonsignificant, the positive t^ value probably indicates a trend for MI to increase for speech produced under stress. Similar results

PAGE 77

57 Tabic 1. Results of the Speech Intensity (SI) Analyses for the Laboratory Stress Experiment; Ail Values are Expressed in dB Re: lmv . Values in Parentheses are the Standard Deviation (SD) and Degrees of Freedom (df) for the t Scores. Speaking Condition Normal Stress Mean Dif(SDj (SD) ference(SD) fc d (d£) A. Mean Intensity (MI) Males (N=7) 67.3(1.7) 68.1(2.4) 0.8(0.7) 2.80(6)* Females (N=5) 65.9(3.3) 65.9(3.1) 0.0(0.8) 0.00(4) Overall (n = 12) 66.7(2.5) 67.2(2.8) 0.5(0.8) 2.07(11) B. Maximum Intensity (MAXI) Males (N=7) 79.1(3.6) 79.6(3.3) 0.4(0.5) 1.96(6) Females (N=5) 78.4(2.9) 78.8(2.9) 0.4(0.5) 1.60(4) Overall (N=12) 78.8(3.2) 79.3(3.0) 0.4(0.5) 2.65(11)* C. Mode of Distribution (MODI) Males (N=7) 73.3(3.8) 76.0(3.6) 2.7(3.0) 2.20(6) Females (N=5) 71.8(6.5) 72.0(6.5) 0.2(1.1) 0.36(4) Overall (N=12) ' 72.7(4.9) 74.3(5.2) 1.7(2.6) 2.17(11) Significant at the 0.05 level. fc .05 = 2.447, df = 6 t 05 = 2.776, df .= 4 t 05 ='2.201, df = 11

PAGE 78

were found for the males alone, the difference of 0.8 dB was found to be significant (t^ = 2.80, df = 6). Conversely, analysis of the female subjects showed no difference between the normal and stress speaking conditions. It would appear that under the present experimental conditions MI v/as not a robust indicator of stress. The results of the analyses of the maximum intensity (MAXI) level are shown in Table 1-B. Analysis of the overall results indicates a statistically significant increase (0.4 dB) in MAXI for the stress condition when it was compared to the normal speaking condition. The maximum intensity produced under the stress condition was 7 9.3 dB, compared to 78. -8 dB for the normal condition. The difference in MAXI between the two conditions was found to be nonsignificant when controlled for the subjects' sex. However, the positive t^ values indicate at least a slight tendency for MAXI to be greater for speaking under stress than for normal speech. The mode of the intensity distribution (MODI) also was analyzed. The results of these analyses may be found in Table 1-C . The 1.7 dB mean difference between the speaking

PAGE 79

conditions was not found to be significant. Moreover, the difference values were found to be 2.7 dB for the males and 0.2 dB for the females, neither of which were significant at the 0.05 level. In each of the analyses MODI was negligibly greater for the stressed speech than for normal speech. Therefore, the tendency — although slight — is for MODI to increase in stressed speech. Finally, the entire intensity distribution (ID) was analyzed for differences between speech produced under stress and normally. The results of these analyses are shown graphically in Figures 3-5.* The distributions were first analyzed by sex, then the data were combined in order to obtain the overall distribution. However, no statistically significant differences were found. Indeed, the general shape of the two distributions (normal and stress) remained fairly constant across conditions. Therefore, analyzing the entire intensity distribution did not discriminate between normal speech and speech produced under stress. In fact, MAXI was found to be the most, powerful stress indicator. The M^XI for the overall data was significantly greater for the stressed speech samples relative to normal speech. The only other significant result was found in MI *The tabulated values for the distributions may be found in Appendix D.

PAGE 80

60 NORMAL STRESS 2 H Uo cc UJ Ql Fiyure INTENSITY (dB re: I MV) Intensity Distribution for the Male Subjec in the Laboratory Stress Experiment.

PAGE 81

b.l NORMAL STRESS x d =0.00 S d = 0.22 t . = 0.00 cl df = 33 (t 05 = 2.035)

PAGE 82

62 Ul U. O Z UJ q: uj

PAGE 83

63 for the males. Therefore, these intensity measures were not extremely indicative' of stress — at least for this experiment . Speaking Fundamental Frequency (SFF) The other major acoustical parameter analyzed was speaking fundamental frequency (SFF). SFF is a measure of the rate of opening and closing of the vocal folds. The mean values of the obtained SFF can be seen in Table 2. The overall mean SFF increased from 166.7 Hz or 39.4 semitones (ST) for the normal speaking condition to 39.7 ST or 169.3 Hz for stress. However, the mean difference between the speaking conditions was not found to be significant. When analyzed by sex, the male subjects' SFF increased from 35.4 ST (127.4 Hz) for normal speech to 36.0 ST (132.0 IIz) for stress, with a mean difference of 0.6 ST (4.6 Hz). The difference for the male subjects was found to be nonsignificant at the 0.05 level. Therefore, SFF for males appears to increase slightly for the stress condition while the females exhibited no change at all in SFF. However, there was a slight tendency for SFF to be higher for speech produced under stress, although these increases were not statistically significant .

PAGE 84

N O 4J N M a o rd Eh CO 01 O TJ -J £ u a -— cu P O CO o O a) c o CJ CO £ tn rj Ql c a, Em W ai a; r: u QJ G a ^ CO H rt) CJ > 4J O C C) in e CJ X £1 QJ 04 Q CO CJ — U N C EC C Q« rj M a qj S l M M p H CO Q *Ecn co p CO E o — 2; a CO U3

PAGE 85

In addition to the mean SFF, the SFF distribution also was analyzed. The resulting relationships may be found in Figures 6-8. The SFF distributions for the two speaking conditions were found to be very similar. The matchedpairs t test for the distributions substantiated this similarity. The distributions were first analyzed by sex, then the data were combined for the overall analysis. However, differences between the normal and stress spanking conditions were not found to be statistically significant. Therefore, the SFF distributions were not very discriminating between normal and stress speech. Spee c h Rate (S R) Speech rate (SR) was defined as the number of syllables produced per second. For the SR analysis the two minute speech samples ware divided into eight 15-second subsamples and alternate subsamples utilized for the computation of the mean and standard deviation of SR. The results of the mean SR analysis may be found in Table 3. As can be seen, the mean value of SR increased from 4.41 syllables/sec Also see Appendix E for the tabulated values for the SFF distributions.

PAGE 86

(.(> 3S u. o liJ O ce LJ 16

PAGE 87

67 NORMAL STRESS UJ 2 u. o z Ul o en UJ ST 20 HZ 52 FUNDAMENTAL FREQUENCY figure 7. Speaking Fundamental Frequency (SFF) Distribution for the Female Subjects in the Laboratory Stress Experiment.

PAGE 88

68 NORMAL cxocoo

PAGE 89

69 Tabic 3. Results of the Speech Rate (SR) Analysis for the Laboratory Stress Experiment; SR Values are Expressed in Syllables Per Second; Values in Parentheses are the Standard Deviations (SD) and the Degrees of Freedom (df ) for the t Scores (N = 12) . Speaking Condition Mean DifNormal (SD) Stress (SD) ference(SD) t d^ df ) Mean 4.41(0.24) 4.42(0.29) 0.01(0.20) 0.17(11) Standard Deviation 0.25(0.15) 0.38(0.17) 0.13(0.21) 2.05(11)

PAGE 90

70 for the normal speech samples to 4.42 syllables/sec for stress. However, this increase was not found to be statistically significant. Similarly, the standard deviation of SR also was analyzed. The results may also be found in Table 3. The SR variability was found to increase for speech produced under stress although not significantly. Although there was a slight tendency for both the SR mean and standard deviation to increase for the stressed speech samples, the increase was not sufficient to differentiate the normal and stress speech. Therefore, neither of the SR measures were very utilitarian in indicating the presence of stress. Speech/Pause Ratio (SPR) The speech/pause ratio (SPR) was the next temporal parameter analyzed. SPR is defined as the ratio of the per cent of speech time to the per cent of pause time. The same subsample segmentation used for SR was also utilized for the SPR analysis. That is, alternate 15 sec subsarnples were used for the computation of the SPR moan and standard deviation. The results of these analyses may be found in Table 4. The analysis of the mean SPR resulted in a nonsignificant difference between speech produced under stress and normally.

PAGE 91

71 Table 4. Results of the Speech/Pause Ratio (SPR) Analysis for th.2 Laboratory Stress Experiment; Values in Parentheses i\re the Standard Deviations (SD) and the Degrees of Freedom (df) for the t Scores (N =12). Speaking Condition Mean DifNormal ( SD) Stress (SD) ference(SD) t d^ df ^ Mean 2.74(0.44) 2.90(0.46) 0.16(0.48) 1.11(11) Standard Deviation 0.47(0.18) 0.58(0.24) 0.11(0.20) 1.82(11)

PAGE 92

7 2 The results of the SPR standard deviation analysis are also shown in Table 4. Similar to the mean, the standard deviation of SPR increased for the stress speech although not significantly. Therefore, SPR did not prove to be an adequate stress indicator. However, both the mean and standard deviation exhibited a slight tendency to increase in speech produced under stress. Time-Energy Distribution (TED) The time-energy distribution (TED) indicates the amount of time a speech signal remains at a particular energy level. An earlier investigation by Johnson (1973) indicated that the tenth (or highest) energy level does not contribute to the discrimination between waveforms. Therefore, in this study only the lower nine energy levels were utilized in the analysis. The TED values were obtained by segmenting the two minute sample into eight 15-second subsamples and computing the mean TED value for every other segment. A summary of the results may be found in Figures 9-11. Initially, the analyses were conducted by sex in order to examine possible differences due to sex; then the data 'See Appendix F for tabulated values of TED.

PAGE 93

73 100 NORMAL STRESS X d = 1.37 S d = 0.92 t d 4.22 of = (t m05 = 2.306) UJ 3E tu_ o Iz UJ o cc I 2 3 4 5 6 7 ENERGY LEVEL Figure S. Time-Energy Distribution (TED) for the Male Subjects in the Laboratory Stress Experiment

PAGE 94

74 100 NORMAL STRESS X d = 1.32 S d = 0.69 t . = 5.42 cl df 8 (t .05 2.306) 30 20 10 I 2 3 4 5 6 7 8 ENERGY LEVEL Figure 10. Time-Energy Distribution (TED) for the Female Subjects in the Laboratory Stress Experiment .

PAGE 95

100 NORMAL STRESS X d 1.3 8 G d = 0.76 t d = 5.14 df = 8 tt.os = 2 306) UJ u. o O I 2 3 4 5 6 7 8 ENERGY LEVEL Figure 11. Overall Time-Energy Distribution (TED) fo: All Subjects in the Laboratory Stress Experiment .

PAGE 96

IC were combined to obtain the overall results. As can be seen, the t scores were all positive and significant at the C.05 alpha level. The _t scores for the males is 4.22, 5.42 for the females and 5.14 for the overall data. The indication is that under stress a speaker will maintain a higher intensity level for a greater portion of his speaking time. This is illustrated in Figures 9-11 by the fact that the stress distribution is elevated relative to the normal distribution. Speech Time/Total Time Ratio (ST/TT) In addition to SPR, the ratio of the speech time to the total sample time (ST/TT) also was analyzed. Top ST/TT measure is the percentage of the total speech sample occupied by speech. The results of the ST/TT analyses are shown in Table 5. The analyses included the mean and standard deviation of ST/TT. The mean ST/TT was found to increase, although not significantly. The mean difference in ST/TT between the normal and stress speaking conditions was only 0.01. Similar results were found for the standard deviation (or variability) of ST/TT; that is, the small increase in variability for stressed speech was not signifi-

PAGE 97

77 Table 5. Results of the Speech Time/Total Time (ST/TT) Ratio Analysis for the Laboratory Stress Experiment; Values in Parentheses are the Standard Deviations (SD) and the Degree of Freedom (df) for the t Scores (M = 12) . Speaking Condition Mean DifNormal(SD) Stress (SD) ference (SD) t d( df ) Mean 0.73(0.03) 0.75(0.03) 0.01(0.04) 0.83(11) Standard Deviation 0.03(0.02) 0.04(0.02) 0.01(0.02) 1.66(11)

PAGE 98

78 In short, ST/TT was not found to be a very powerful indicator of stress in this experiment. However, there was a tendency for both the mean and standard deviation of ST/TT to increase slightly for speech produced under stress. Speech Bursts and Pauses Two other temporal measures obtained were the number of speech bursts and pauses. The results of these analyses may be found in Table 6. As can be seen, the number of speech bursts as well as the number of pauses was greater for the stress speech samples. However, the increase was inconsequential (0.1) for both measures and not statistically s ignificant Pis fluency The final measure analyzed in this experiment was that of disfluency. This measure was obtained utilizing the total number of disfluences in the two minute speech sample; The mean values for the two speaking conditions are shown in Table 7. A:; can bo seen, the numbnr of d Lsf luencios wars greater for the stressed speech (7.7) than for the normal (5.2). in addition, the 2.5 increase was found to be stati.' tically significant. Therefore, the affect of stress on

PAGE 99

7 9 Table 6. Results of the Speech Bursts and Pause Analyses for the Laboratory Stress Experiment; Value in Parentheses is the Degrees of Freedom (df) for the t Scores (W = 12) . Speaking Condition Mean t (d*) Normal Stress Difference d v " ' A. Speech Bursts Mean 40.0 Standard Deviation 5.7 40.2 6.6 0.1 3.2 0.10(11) B. Pauses Mean 40.6 Standard Deviation 5.7 40.7 6.6 0.1 3.4 0.10(11)

PAGE 100

30 Table 7. Analysis Results of the Number of Disfluencies for the laboratory Stress Experiment; Value in Parentheses is the Degrees of Freedom (df) for the t Score (M = 12) . Speaking Condition Mean Normal Stress Difference t d ( df ) Mean 5.2 7.7 2.5 2.36(11) Standard Deviation 3.1 2.9 *Significant at the 0.05 level.

PAGE 101

81 the level of speech disfluency is substantial enough to suggest that this measure might be utilized as a vocal indicator of stress.

PAGE 102

CHAPTER IV RESULTS OF SITUATIONAL STRESS EXPERIMENT The purpose for conducting the second experiment was to obtain information under conditions of a different and greater stress. In this experiment, the stress experienced by the subjects was induced by having them deliver a public speech to .a group of their peers. For the majority of the subjects this was the first exposure they had experienced to a situation of this type and they naturally were anxious. Nevertheless, the behaviors of only those individuals scoring 13 or higher on the MAACL anxiety scale were studied . The same values were obtained and analyzed in this experiment as in the previous investigation. That is, within the acoustic domain they included maximum, mean, mode and distribution of speech intensity (SI) as well as speaking fundamental frequency (SFF) mean and distribution. The measures within the temporal domain included mean and standard deviation of speech rate (SR), speech time/total time (ST/TT) ratio, and speech/pause ratio (SPR) . Finally, 82

PAGE 103

83 also included were time-energy distribution (TED) measures and the mean number of speech bursts and pauses, as well as the number of disf luencies . As with the previous experiment, the statistical procedure utilized was a matched-pairs, two-tailed _t_ test. It was used in order to test the null hypothesis — i.e., that there is no difference between the acoustical/temporal measures for speech produced under conditions of stress and for those produced when there is no stress . Speech Intensity (Si) Of the three intensity characteristics analyzed, the first to be reported will be mean intensity vKI) ieve_L, these results may be found in Table 8-A. It may be seen, MI was greater for normal speech than for speech produced under stress both by sex and for the subjects pooled. These relationships are indicated by the negative mean difference values. The reduction in MI for the stressed speech ranged from 3.3 dB for the females to 6.1 dD for the males. The trend was sufficiently consistent to produce significant _t scores. The implication of these data is that, for conditions of situational stress, mean speech intensity will be reduced from the normal.

PAGE 104

84 Table 8. Results of the Speech Intensity (SI) Analyses for the Situational Stress Experiment; All Values are Expressed in dB Re: lmv. Values in Parentheses are the Standard Deviations (SD) and the Degrees of Freedom (df) for the t Scores. Speaking Condition Normal S tress (SD) Mean Difference (SD) t d (df) A. Mean Intensity (MI) Males (N = 10) 68.4(2.5) 62.3(4.0) -6.1(3.5) -5.23(9)* Females(N=7) 67.6(1.8) 64.3(1.8) -3.3(2.2) -3.67(6)* Overall(N=17) 68.1(2.2) 63.1(3.4) -4.9(3.3) -5.94(16)* B. Maximum Intensity (MAXI) Males (N = 10) 81.6(1.2) 73.7(3.1) -7.9(2.7) -8.78(9)* Females (N=7) 81.4(1.1) 76.4<2.4) -5,0(3.1) -3.95(6)* Overall (N = 17) 81.5(1.1) 74.8(3.0) -6.7(3.1) -8.65(16)* C. Mode of Distribution (MODI) Males (N=10) 74.6(4.9) 66.4(7.5) -8 . 2 (5 . 7 ) -4 . 32 (9) * Females (N=7) 73.4(4.9) 70.0(3.6) -3.4(4.6) -1.81(6) Overall (N = 17) 74.1(4.8) 67.9(6.3) -6.2(5.7) -4.35(16)* fSignif icant at the 0.05 level. 05 2.262, df = 9 t Q5 = 2.447, df = 6 t_ 05 = 2.120, df = 16

PAGE 105

The second intensity measure to be analyzed was maximum intensity (MAXI) . The MAXI results also may be found in Table 8-B. MAXI was found to decrease by 7.9 dB for the males, 5.0 dB for the females and 6 . 7 dB for the combined results and all three differences between the two speaking conditions were found to be statistically significant. These results would appear to indicate that, under conditions of stress, the maximum speech intensity subjects produce tends to be somewhat less than it does when they are speaking under neutral conditions. The third intensity metric studied was the mode of the intensity distribution (MODI). See Table 8-C for the results of tiiis analysis. As with the previous measures, when compared to the normal MODI was found to decrease for speech produced under stress conditions. The males exhibited the greatest decrease (8.2 dB) and the females, the least (3.4 dB) . The overall results, obtained by combining the results for the males and females, showed a decrease of 6.2 dB. However, only the results for the males and the overall analysis were found to be statistically significant. Nevertheless, the tendency for MODI was to decrease for stressed speech.

PAGE 106

06 As a follow-up, the overall intensity distribution also was analyzed to provide more insight into these relationships. The results of these evaluations may be seen in Figures 12-14. Here the data are presented by sex and also combining all the subjects. The statistical analyses of the distributions indicated little or no difference between the two speaking conditions (normal end stress). However, some weight can be obtained by consideration of these distributions. For example, the first impression is of a "shift" in the stress distribution (relative to the normal). It was found that for intensity levels less than approximately 65-75 dB the stress distribution was greater than the normal. Conversely, above the 65-75 dB level, the distribution for the stress condition is less than the normal distribution. However, the "shift" was such that the differences between the distributions at the lower intensities counter-balanced the differences at the higher intensities resulting in a nonsignificant mean difference between the stress and normal intensity distributions. The cited results appear to indicate that the measures of the mean, maximum or mode of the intensity distribution The tabulated values for the distributions may be found in Appendix D.

PAGE 107

a 7 NORMAL STRESS UJ 2 U. o hz UJ ce UJ a. x d = 0.00 S d =2.20 t d = 0.00 df = 3 3 (t nK = 2.035) INTENSITY (dB re: I MV) Figure 12. Intensity Distribution for the Male Subjects in the Situational Stress Experiment .

PAGE 108

UEJ UJ 2 u. O Ul o UJ

PAGE 109

6 o H 2L UJ o UJ a.

PAGE 110

90 would be better stress indicators than the entire distribution . Speaking Fundamental Frequency (SFF) As with the first experiment, the SFF parameters selected for analysis were the mean and SFF distribution. The results of these several SFF analyses may be found in Table 9. It should be noted that the data are tabulated in both semitones (ST) and hertz (Hz) but the analyses will be discussed only in terms of ST because of the geometric nature of SFF. As with the other measures, the data were analyzed first by sex then combined to obtain the overall results. Examination of the table will demons tra te that the males and the overall analyses showed significant increases in SFF for speech produced under stress. The females also showed an increase in f Q , although it was not significantly different. That is, the increase for the females was only 0.1 ST as compared to 1.3 ST for the males and the overall increase of 0.8 ST. The SFF distribution also was analyzed for potential dif forenoon between speecn produced under neutral and stress conditions. As with the mean, these results are

PAGE 111

J I rH

PAGE 112

92 shown in Figures 15-17* As can bo seen, no significant difference was found between the two distributions for any of the analyses. Indeed, in each case, the analysis resulted in a _t score of 0.00 and the general shape of the distributions for the different speaking conditions appear to be very similar. To summarize, the mean SFF was found to be more important than the SFF distribution for determining the presence of stress. The mean SFF was found to increase for both the males and the females although only the increase for the males was significant. Furthermore, the overall SFF was found to be significantly greater than normal for the stress condition. However, the SFF distributions for each sex as well as the males and females combined showed no difference between the two speaking conditions. Spe ech Rate (SR) As per the protocols utilized in the previous experiment, the two minute speech samples were divided into 15 second subsamples for speech rate (SR) analysis. The first, third, fifth and seventh subsamples were utilized to compute the mean and standard deviation of SK. Table 10 *See Appendix E for the tabulated values for the SFF dis tributors .

PAGE 113

93 IxJ H u. o H 2: UJ o DC IU Q. 16

PAGE 114

UJ 2 u. o 94 NORMAL STRESS ST20

PAGE 115

95

PAGE 116

Mean 96 Table 10. Results of the Speech Rate (SR) Analysis for the Situational Stress Experiment; SR Values Are Expressed in Syllables Per Second; Values in Parentheses Are the Standard Deviations (SD) and the Degrees of Freedom (df) for the t Scores (N =17). Speaking Condition Normal Stress Mean Dif(SD) (SD) ference(SD) fc d( df ) 4.5(0.5) 4.2(0.5) -0.3(0.6) -2.00(16) Standard Deviation 0.3(0.2) 0.6(0.3) 0.3(0.2) 6.00(16) *Significant at the 0.05 level,

PAGE 117

97 provides the mean values and standard deviations of SR under the two speaking conditions studied. It also provides the mean difference (stress minus normal) between these conditions. The SR was found to be less for the stressed speech (4.2) than the normal (4.5); however, the decrease was not enough to be statistically significant. The variability (as measured by the standard deviation) for the normal speaking condition was found to be 0.3 syllables/sec as compared to 0.6 syllables/sec for the stress samples. This difference of 0.3 resulted in a significant increase in the variability of SR for speech under stress conditions compared to normal. In anv case, no significant change was found for mean SR between the two speaking conditions. However, the data suggest that when stressed, individuals will vary their SR more than they will under normal circumstances. That is, under stress they fluctuate SR to greater extremes compared to normal, while maintaining the same mean SR. Spoech/Pause Ratio (S PR) The speech/pause ratio (SPR) is defined as the ratio of total speech time to the pause time in the overall sample, As with speech rate, SPR was calculated using alternate 15

PAGE 118

98 sec subsamples extracted from the two minute speech sample. The mean, standard deviation and mean difference in SPR for each speaking condition are shown in Table 11. Both the SPR mean and standard deviation can be seen to increase significantly for the stress speech samples; the mean by 1.32 and the standard deviation, 0.87. In short, these data imply that, under conditions of stress, the average amount of speech for a given time period will increase relative to the amount of pause time. Furthermore, the variability of SPR also can be expected to increase . T ime-Energy Distributi on (TED) As has been stated, the time-energy distribution (TED) is a measure of the amount of time speech signal energy reaches a specified level. That is, the energy range of the signal is divided into ten equal levels and the percentage of the total sample time the signal remains in each level is determined. However, only the first nine levels are incorporated in the data analysis of the present study. The decision to eliminate the tenth level is based on earlier research by Johnson (1970). In his speaker identification study, Johnson found that the tenth energy level did not

PAGE 119

99 Table 11. Results of the Speech/Pause Ratio (SPR) Analysis for the Situational Stress Experiment; Values in Parentheses are the Standard Deviations (SD) and the Degrees of Freedom (df) for the t Scores (N = 17) . Speaking Condition Normal Stress Mean Dif(SD) (SD) ference (SD) t d (df) Mean 3.13(0.95) 4.44(1.53) 1.32(1.06) 4.98(16) Standard Deviation 0.61(0.39) 1.48(0.83) 0.87(0.89) 3.91(16) *Significant at the 0.05 level

PAGE 120

100 contribute to the discrimination of different speech samples. That is, once a discrimination level had been reached utilizing the first nine energy levels, the addition of the tenth level did not improve the ability of TED to differentiate speech samples. The results of the TED analyses are presented in Figure 18.-20.* A relatively strong and consistent relationship can be seen to exist between the speaking condition and TED. Specifically, the TED was found to increase significantly for speech produced under stress when the samples were contrasted to normal speech. This relationship held true for each of the sex subgroups as well as for the pooled groups . These data indicate that, under conditions of stress, a speaker will produce speech at higher energy levels for a greater proportion of time than they would if they were speaking normally. Although, under stress, the higher intensity levels are produced a greater amount of the time, the overall intensity level is reduced from the normal (as reported earler in the speaking intensity section). *The tabulated values for the TED may be found in Appendix F.

PAGE 121

101 100 —

PAGE 122

102 100 NORMAL STRESS 90 80 X d = 6.12 S d = 1.71 t , 10.12 o df = 8 (t_ 05 = 2.306) 4 5 ENERGY LEVEL Figure 19. Tiwe-Energy Distribution (TED) for the Female Subjects in the Situational Str« Experiment ,

PAGE 123

103 100 NORMAL STRESS UJ 2 u_ O Z UJ o a: UJ a. 40 x n

PAGE 124

J 04 Speech Time /Total Time Ratio (ST/TT) Another temporal ratio analyzed was that of the speech time to the total time of the sample (ST/TT). Although this metric is related to the speech/pause ratio (SPR) they are different measures. SPR compares the speech time to the pause time whereas ST/TT contrasted the speech time to the total sample time. Both the mean and standard deviation of ST/TT were analyzed and tine results may be found in Table 12. As indicated in Table 12, the mean ST/TT was 0.05 greater for the stress speech samples than for the normal samples. Similarly, the standard deviation also increased, although by only 0.02. The increase in mean ST/TT was found to be statistically significant, while the slight increase in standard deviation was not significant. Therefore, it can be concluded that an individual speaking under stress will generally tend to spend a greater percentage of the total time speaking than he or she would normally. Furthermore, based on the SPR and ST/TT results, it can be seen that the speech time increases as compared to both the pauses and the total time of the speech samples.

PAGE 125

105 Table 12. Results of the Speech Time/Total Time (ST/TT) Ratio Analysis for the Situational Stress Experiment; Values in Parentheses are the Standard Deviations (SD) and the Degrees of Freedom (df) for the t Scores (N = 17) . Speaking Condition Normal Stress Mean Dif(SDJ ference(SD) h d (df j Mean 0.74(0.05) 0.79(0.06) 0.05(0.05) 4.00(16) Standard Deviation 0.04(0.02) 0.06(0.04) 0.02(0.04) 2.00(16) *Significant at the 0.05 level.

PAGE 126

106 Speech Bur sts and Pa uses Two additional temporal parameters analyzed were the actual number of speech bursts and pauses. The results of these analyses may be found in Table 13. As would be expected, these measures are related to both SPR and ST/TT. Consideration of these data will reveal that when an individual is speaking under conditions that induce stress, his or her speech apparently reflects a reduced number of speech bursts and pauses. As can be seen in Table 13, both the bursts and the pauses decreased significantly for the stress samples as compared to the normal. Combining these results with those for SPR and St/TT the speech pattern under stress might be described as follows. An individual will produce an extended utterance, pause briefly, then continue speaking in another extended speech segment. In this manner, the individual produces fewer speech bursts (of longer duration) and pauses in addition to increasing the speech time as compared to the pause time and total sample time. Disf luen c_y The final measure analyzed in this experiment was that of disfluency. In this case, the number of times the

PAGE 127

107 Table 13. Results of the Speech Bursts and Pause Analyses for the Situational Stress Experiment; Value in Parentheses is the Degrees of Freedom (df) for the t Scores (N = 17) . Speaking Condition Normal Stres: Mean «« t, df) Difference d v ' A.

PAGE 128

108 speaker was disfluent was counted for the entire two minute sample. These data can be found in Table 14. The mean number of disfluencies was greater for speech produced under conditions of stress (10.2) than it was for the neutral sample (6.8). However, the difference between these conditions was not found to be significant. These findings indicate that this measure of disfluency may not be as useful an indicator of stress as had been anticipated. However, the non-significance of the increase in disfluency may have been due to the speakers 1 familiarity with stress speech sample (i.e., their public speeches). Since the subjects had practiced the particular sample they tended to be more fluent than if they had not practiced. Therefore, disfluencies may not indicate stress when the speaker produces a prepared speech sample.

PAGE 129

109 Table 14. Analysis Results of the Number of Disfluencies for the Situational Stress Experiment; Value in Parentheses is the Degrees of Freedom (df) for the t Score (N = 17). Speaking Condition Mean Stress Difference d [ ' Normal Mean 6.8 Standard Deviation 5.1 10.2 5.5 3.2 6.5 1.97(16)

PAGE 130

CHAPTER V DISCUSSION As would be expected, the results of the two current studies indicate that measurable acoustical/temporal changes do occur in speech produced under stress — that is, when it is compared to normal, nonstress speech. Based on the present data, however, some of the observed vocal changes were not in the direction that had been anticipated. Furthermore, the magnitude of the differences between the normal and stress speech samples appears to be a function of the type of stress. For example, more of the measures analyzed for the situational stress experiment showed significant change than did those in the laboratory experiment . Laboratory Stress Experiment A summary of the results of the laboratory stress experiment may be found in Table 15. These results are based on the overall mean values for each of the parameters analyzed. As can be seen from the table, the mean values 110

PAGE 131

Ill Table 15. Summary of Results for the Laboratory Stress Experiment Based on Overall Mean Values for Each Parameter. Speech Parameter Direction of Change (Stress re: Normal) A. Acoustical Speech Intensity (SI) Speaking Fundamental Frequency (SFF) B. Temporal Speech Rate (SR) Speech/Pause Ratio (SPR) Time-Energy Distribution (TED) Speech Time/Total Time Ratio (ST/TT) Speech Bursts Pauses C. Disfluency + = Change in direction toward increase = Change in direction toward decrease * = Statistically significant at the 0.05 level

PAGE 132

112 for all the parameters analyzed in this experiment showed change--specif ically increases — for the stressed speech when it was compared to the neutral condition. However, many of these changes were not statistically significant. For example, the maximum intensity (MAXI) was the only intensity measure to increase significantly, although the general trend was for all intensity measures, i.e., the maximum, mean, mode and distribution, to increase at least slightly. Similarly, the mean and distribution of speaking fundamental frequency (SFF) both increased slightly under stress conditions. However, these increases were not found to be of statistical significance either. Examination of Table 15 also will reveal that, within the temporal domain, the only parameter to change significantly for the stress speech condition was the time-energy distribution (TED) . The remaining temporal parameters of speech rate (SR) , speech/pause ratio (SPR) , speech time/ total time ratio (ST/TT) , speech bursts, and pauses were found to be greater than normal under conditions of stress, although not significantly so. Finally, the disfluency metric increased significantly when the speech samples were produced under stress.

PAGE 133

113 In addition to increases in the mean values of those parameters studied, variability also increased. Quite often a large variability (as measured by the standard deviation) resulted in the differences between two means not being significant. That is, the measured changes in the acoustical/ temporal parameters included in this research turned out to be inconsistent across individuals. Moreover, the stressing effects of electro-shock, in this particular experiment, did not appear to alter the subjects' speech patterns to any significant degree. Therefore, few of the parameters appear to show compelling promise as potential indicators of stress. On the other hand, it is possible that electric shock did not stress the subjects studied to a reasonably high degree and that application of other types of stressors would result in conditions that would lead to the uncovering of the anticipated relationships. For example, most of the trends observed in these two experiments were found to be consistent for both. To illustrate, speaking fundamental frequency, speech/pause ratio and the speech time/total time ratio were found to increase (and significantly) in the situational stress experiment. Thus, it would appear that the trends found in the laboratory stress experiment may be in the right direction, and that the stressor was not severe enough to move them to significant levels.

PAGE 134

114 Situational Stress Experiment: The results of the situational stress experiment are summarized in Table 16. As can be seen, many of the relationships between speech and emotional stress were found to be significant. That is, subjects' speech was significantly modified under the stress of a public speaking task. Significant differences between the two speaking conditions (i.e., neutral/stress) were found both acoustically and temporally . Certain of the results in the acoustic domain are of particular interest. For example, the results show a significant decrease in speaking intensity and an increase in mean speaking fundamental frequency. The decrease in intensity was consistent for the mean, maximum and mode. Within the temporal domain, the speech/pause ratio, time-energy distribution and -speech time/total time ratio all were found to be significantly greater for the stress speech samples than for the normal samples. Conversely, the results for speech rate, the number of bursts and pauses indicated a decrease relative to neutral speech. However, the decrease in speech rate was not significant. In addition to these findings, it can be noted that the speech changes resulting from the public speaking situation

PAGE 135

115 Table 16. Summary of Results for the Situational Stress Experiment Based on Overall Mean Values for Each Parameter. Direction of Change Speech Parameter . (Stress re: Normal), A. Acoustical Speech Intensity (SI) Speaking Fundamental Frequency (SFF) B. Temporal Speech Rate (SR) Speech/Pause Ratio (SPR) Time-Energy Distribution (TED) + Speech Time/Total Time Ratio (ST/TT) +^ Speech Bursts Pauses C. Dis fluency * * + = Change in direction toward increase = Change in direction toward decrease * = Statistically significant at the 0.05 level

PAGE 136

116 were more complex than that for laboratory stress. For example, speech intensity decreased with a simultaneous increase in SFF . This relationship, between intensity and SFF, is contrary to accepted theory that vocal intensity and SFF are positively correlated (Issihiki, 1959; Ladefoged and McKinney, 1963). However, in the present case, the differences are behaviorally based. Thus, it is quite possible that the behavioral effects were powerful enough to override the natural physiological relationships. Further, it was found that the number of speech bursts decreased simultaneously with a significant increase in both the speech/pause and speech time/total time ratios. These particular temporal findings would suggest that the subjects spoke in extended utterances. That is, they appeared to produce fewer speech bursts and the overall duration for these bursts was greater in this case than for neutral speech. The number of pauses also were found to be fewer for speech produced under stress than for normal speech; a finding that is consistent with that for increases in speech/pause ratios. in short, tasks based on situational stress resulted in speech that showed: (1) lower intensity, (2) higher speaking fundamental frequency, (3) slightly slower rates, (4) fewer

PAGE 137

117 (but longer) speech bursts, and (5) fewer pauses. However, as with the laboratory stress experiment, the large variability for several of the parameters suggests that the resultant parameter changes were not consistent for all the s ub j e c t s . It should be noted that, for the laboratory stress experiment, both speaking conditions were recorded in an IAC sound treated room and that the same speech materials ("An Apology for Idlers") were utilized. However, for the situational stress experiment, the stress samples were recorded in an open classroom and consisted of different semiextemporaneous speech materials while the normal speech samples were recorded in the IAC room and consisted of a reading of "An Apology for Idlers." However, when the results of the two experiments are compared (see Tables 15 and 16), the general agreement (in direction usually) of the speech changes becomes obvious. The only disparities might be in speech intensity, number of speech bursts and pauses where these parameters decreased for the situational stress experiment and tended to increase for laboratory stress. However, the increases were only slight and were not found to be significant. Nevertheless, there are more similarities between the results of the two experiments than there are

PAGE 138

118 differences (see Table 17). It is on this basis that a kind of cross-verification of the results occurs. That is, the data from the first study cross-validate the second and the data from the second tend to give credence to the trends in the first (which, otherwise, are non-significant). Comparison to Previ ou s Studies; SFF Prior to the present study, the acoustic parameter most frequently investigated as a potential correlate of stress effects on speech has been fundamental frequency (f Q ) (Fairbanks and Pronovost, 1939; Bonner, 1943; Becker et_al . , I960; Williams and Stevens, 1969 and 1972). Although several of the previous studies utilized actors simulating various emotions, certain comparisons can be made between these investigations and the present data. For instance, Fairbanks and Pronovost (1939) found that for the emotions they studied (contempt, anger, fear, grief and indifference), fear was reflected by the highest SFF. Similar results were reported by Williams and Stevens (1972) using the emotions sorrow, fear, anger and a neutral speaking condition. r a ti-,-,tfn,ir had the second highest Williams and Stevens found that tear naa mean SFF, with anger being higher. In an earlier (1969) study, Williams and Stevens also found that pilots and

PAGE 139

119 Tabic 17. Comparison of the Overall Results for the Laboratory and Situational Stress Experiments Based on Mean Values for Each Parameter. Speech Parameter Direction of Change (Stress re: Normal) Laboratory Situational Stress Stress Acoustic Speech Intensity (SI) + Speaking Fundamental Frequency (SFF) + Temporal Speech Rate (SR) + Speech/Pause Ratio (SPR) + Time-Energy Distribution (TED) + Speech Time/Total Time Ratio (ST/TT) + Speech Bursts + Pauses + C. Disfluency + + Change in direction toward increase = Change in direction toward decrease * = Statistically significant at the 0.05 level

PAGE 140

12 control tower operators increased f Q under stress conditions (presumably, fear). On the other hand, Hecker et_al. (1968) reported inconsistent changes in f Q for their subjects who were experiencing stress. For the present study, mean SFF was found to increase — for the stressed speech samples — in both experiments. When the results of the above studies are combined relative to the f effects of fear/anxiety/ stress, the pattern seems to indicate that a speaker's SFF will increase from the norm for conditions of stress. C omparison to Previous Studies: Inten sity. Two earlier groups of investigators (Friedhoff c t al., 1964; Hecker et al., 196b) have examined the relationship between speaking intensity and stress. Friedhoff et_al. found that average intensity increased when subjects lied (and presumably were speaking under stress). Hecker et_al. reported small or inconsistent differences in intensity levels between their control and stress speaking conditions; however, one of their subjects did exhibit an increase of 2.0 dB for the stress condition. By comparison, the data from the current laboratory stress experiment show a slight, but nonsignificant, increase in intensity for stress related speech productions. However, for the situational stress

PAGE 141

121 experiment, these same measures manifest a significant decrease in intensity for the stressed speech. The difference in these results may have been due to a difference in the types of stress applied to subjects in the two experiments. Furtheremore , the discrepancy between the present experiments and earlier studies may be explained by the fact that Friedhoff et al . (1964) analyzed only one word and Iiecker e t al . (1960) utilized "certain portions of the test phrase" (p. 995) whereas the entire speech sample was analyzed in the present case. Comparison to Previous S t udies: Temporal Temporal analysis of speech — as it relates to emotions-was extensively studied by Fairbanks and Hoaglin (1941) and also by Williams and Stevens (1972) . in addition, Hecker e t al . (1960) refer to changes in the "duration of phonetic segments" (p. 1001) as potential articulatory manifestations of stress. Fairbanks and Hoaglin (1941) found that conditions of (emotional) fear resulted in the second highest mean overall rate of 202 words per minute, followed closely by anger. However, Williams and Stevens (1972) report that when compared to their neutral speaking condition, the fear condition resulted in a slower speech

PAGE 142

122 rate (i.e., 3.80 syllables per second). Fairbanks and Iloaglin (1941) also studied the mean number of phonations and pauses for the simulated emotions of contempt, anger, fear, grief and indifference. They report that fear had the second lowest number of phonations and pauses. It is difficult to compare the rate measures obtained by Fairbanks and Iloaglin (1941) to those of Williams and Stevens (1972) due to their differential use of units of measure. Moreover, it is difficult also to compare the results from the present study to these experiments as the data here are obtained by different means; anyway the present results do not seem conclusive. For example, speech rate increased for the laboratory experiment, but decreased in the public speaking experiment, although neither change was found to be significant. However, the number of speech bursts and pauses were found to be significantly less for the situational stress speech than for the control (nonstress) speech, a finding which is in agreement with those of Fairbanks and Iloaglin. For the laboratory stress experiment, the number of speech bursts and pauses tended to increase for stressed speech; nevertheless, the increase was only slight and not found to be significant.

PAGE 143

123 Conclusions The findings of this research may be viewed yet differently; that is, they show that, of the acoustical/ temporal measures analyzed, 68.8% changed significantly between the two speaking conditions in the situational stress experiment, whereas only 18.8% changed significantly in the laboratory stress experiment. These relationships may be explained, in part, by a differential subject response to the two stressors. Specifically, it will be remembered that the definition proposed in Chapter I was that stress is a psychological state that is a response to a p j! rce J; vjdJ^reat and will be accompanied by speckle emotions (e.g., fear, anxiety, anger). The major difference between the two present experiments may have been embedded in the threat perceived by the subjects. In the laboratory experiment, the subjects had no emotional involvement with the speaking task; they were free to end the experiment at any time and the threat was primarily a physiological one. However, in the situational stress experiment, the subjects probably were under more emotional pressure and undoubtedly perceived a greater threat-i.c, the possibility of failure or not doing well in front of a peer

PAGE 144

124 group. Therefore, it can be argued that the subjects in the situational stress experiment perceived a substantially greater threat and the effects of this (greater) stressor were manifested in their speech. Thus, it would appear that the level/type of stress must be such that the perceived threat exceeds a particular threshold before the stress is reflected in an individual's speech. Based on the results of the acoustical/temporal analyses conducted in the present study, certain conclusions can be drawn. These conclusions relate to the general speech/voice patterns of an individual who is experiencing stress. Specifically, his or her patterns might be characterized as exhibiting: 1. a generally decreased overall intensity, 2. a generally increased speaking fundamental frequency, 3. somewhat decreased speaking rates, 4 longer speech bursts within an utterance; that is, fewer (but longer) speech bursts and fewer pauses. Although the above characteristics may describe the overall pattern for speech under stress, individual subject variability may be expected. That is, the speech changes that resulted from the effects of stress were not

PAGE 145

125 necessarily consistent for all subjects. For one person, an increase in a particular speech parameter might indicate the presence of stress, whereas in another person a decrease in the same parameter may be equally indicative. Therefore, "speaking stress" profiles may have to be established for an individual if attempts are to be made subsequently to determine if that person is experiencing stress and only speech samples are available as indicators.

PAGE 146

APPENDICES

PAGE 147

APPENDIX A MULTIPLE AFFECT ADJECTIVE CHECK LIST

PAGE 148

MULTIPLE AFFECT ADJECTIVE CHECK LIST in general form By Marvin Zuckerman and Bernard Lubin Name Age Sex Date Highest grade completed in school DIRECTIONS: On this sheet you will find words which describe different kinds of moods and feelings. Mark an /x7 i n the boxes beside the words which describe how you generally feel . Some of the words may sound alike, but we want you to check all the words that d escribe your feelings. Work rapidly. 123

PAGE 149

129 i £7

PAGE 150

APPENDIX B SUBJECT INFORMED CONSENT FORM

PAGE 151

SUBJECT INFORMED CONSENT FORM University of Florida Department: Name of Subject: Address : Title of Project: Name of Investigator and Academic Title: Project Number (assigned by the Departmental Committee) : I, the undersigned, after having read the attached instructions, do understand the general nature of the investigation entitled above. Also, the procedures to be followed have been described to me by the investigator whose name is signed below. I agree to participate in this study, or to have my minor child, or ward, whose name, is , — , — _ participate in this study. I have been informed of the right to withdraw at any stage of the project. (Subject's Signature) (Date) (Signature of parent or guar(Date) dian, if subject is a minor (where applicable) ) I, ths undersigned, have described to the volunteer, whose signature is given above, the general nature of this study and the procedures to be followed. (Investigator's Signature) (Date) (Witness) " (Date) Written instructions to the subject are attached 131

PAGE 152

APPENDIX C INSTRUCTIONS TO SUBJECTS

PAGE 153

INSTRUCTIONS TO SUBJECTS Please read the attached prose passage to familiarize yourself with its contents. After you have gone over it you will be asked to read it twice. First in a normal conversational style; secondly, you will be subjected to stress by a random presentation of mild electric shock, to your index finger. The entire procedure should take approximately twenty minutes. If at anytime, during the experiment, you wish to terminate, please feel free to do so by informing the experimenter. The shock level used will cause minimal discomfort. Stress will actually be induced by the randomness of presentation . No monetary compensations will be made for participation in this experiment. Feel free to ask any questions you may have . 133

PAGE 154

134 Adapted from: AN APOLOGY FOR IDLERS By Robert Louis Stevenson If you look back on your own education, I am sure it will not be the full, vivid, hours of truancy that you regret. You would rather cancel out some of the lack-luster periods between sleep and waking that you experienced in school. For my own part, I have attended a good many lectures in my time— I still remember that the spinning of a top is a case of kinetic stability. But though I would not willingly part with such scraps of science, I do not set the same store in them as by certain other odds and ends that I came upon in the open street while I was playing truant . Extreme busyness, whether at school or college, church or market, is a symptom of deficient vitality. A faculty for idleness implies a catholic appetite and a strong sense of personal identity. There are a sort of dead-alive, hackneyed people about, who are scarcely conscious of living except in the exercise of some conventional occupation. Bring these follows into the country, or set. them on board ship, and you will see how they pine for their desk or their study. They have no curiosity; they cannot give themselves over to random provocations nor do they take pleasure in the

PAGE 155

135 exercise of their faculties for its own sniic. Unless necessity lays about them with a stick, they will even stand still. it is no good speaking to such folk. They cannot be idle; their nature is not generous enough. They pass those hours, which are not dedicated to furious toiling in the gold-mill, in a sort of coma. When they do not require to go to the office, when they are not hungry or have no mind to drink, the whole breathing world is a blank to them. If they have to wait an hour or so for a train, they fall into a stupid trance with their eyes open. To see them you would suppose there was nothing to look at and no one to speak with. You would imagine they were hypnotized or frozen. Yet, very possibly they are hard workers in their own way, and have good eyesight for a flaw in a deed or a turn of the market. They have been to school and college, but during all that time they had their eye only on their grades. They have gone about in the world and mixed with clever people, but all the time they were thinking only of their own affairs. As if a man's soul were not too small to begin with, they have dwarfed and narrowed theirs by a life of all work and no play. Here they are forty, with a listless attention, a mind vacant of all material of amusement, and not one thought to rub against another while

PAGE 156

13 CS they wait for that train. Before he grew up he might have clambered on boxes. When he was twenty he would have staredat the girls. But now the pipe is smoked out, the snuffbox empty, and my gentleman sits bolt upright upon a bench, with vacant eyes. This does not appeal to me as being a "Success in Life." But it is not only the person himself who suffers from his busy habits, but his wife and children, his friends and relations, and even the very people he sits with in a railway carriage or a bus. Perpetual devotion to what a man calls his "business" is only to be sustained by perpetual neglect of many other things. In fact, it is not by any means certain that a man's "business" is the most important thing he has to do.

PAGE 157

APPENDIX D TABULATED VALUES FOR THE INTENSITY DISTRIBUTIONS

PAGE 158

Tabic D-l. intensity Distribution Values for the Male Subjects in the Laboratory Stress Experiment; Intensity Values are Expressed in dB re : lmv . — '

PAGE 159

l3g Table D-2. Intensity Distribution Values lor the Female Subjects in the Laboratory Stress Experiment; Intensity Values are Expressed in uB re : lmv . Intensity (riB) Speaking Condition Normal Stress Difference 50 51 52 53 54 55 56 57 58 59 60 6] 62 63 64 65 66 67 68 69 70 7 1 72 73 74 7 5 76 7 7 78 79 80 81 82 83 . 6% 2.4 1.9 2.8 2.3 2.6 2.7 2.7 2.9 3.2 3.4 3.5 3.5 3.8 3.6 4.1 4.3 4.0 2.9 2.6 2.7 2.1 1.7 1.5 1.0 . 3 0.0 . 6%

PAGE 160

140 Table D-3. Intensity Distribution Values for All Subjects in the Laboratory Stress Experiment; Intensity Values Are Expressed in dB rc;lmv. Intensity (dp) Speaking Condition Normal Stress Difference 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 6 9 7 71 72 73 74 7 5 76 7 7 78 79 80 8 1 8 2 83 0.7% 1.8 1.8 2.3 2.0 2.1 2.3 2.5 2.6 2.9 3.1 1 2 4 6 9 2 5 7 5.3 5.4 4.7 3.9 3.4 3.1 0.7% 1.7 1.8 2.0 2.0 2.2 2.3 2.4 2.4 2.6 2.8 2.9 3.2 3.4 3.5 3.7 3.7 4.2 .5 .5 .8 .2 .2 .7 .1 .7 .4 .1 .6 .4 .2 .7 0.2 0.0 4 4 4 5 5 4 4 3 3 3 2 2 2 1 0.7 0.1 0.0% -0.1 0.0 -0.3 0.0 0.1 0.0 -0.1 -0.2 -0.3 -0.3 -0.2 0.0 0.0 -0.1 -0.2 -0.3 0.0 0.0 -0.2 -0.2 -0.1 -0.2 0.0 0.2 0.3 0.3 0.1 0.1 0.3 0.5 0.7 0.5 0.1 _ 0.0.1 0.24 Mean S.D. t d = 0.24, df = 33 (t 05 = 2.035)

PAGE 161

141 Table D-4. Intensity Distribution Values for the Male Subjects in the Situational Stress Experiment; Intensity Values are Expressed in clB rerlrav.

PAGE 162

142 Table D-5. Intensity Distribution Values tor the Female Subjects in the Situational Stress Exper iment ; Intensity Values are Expressed in dB re : lmv.

PAGE 163

143 Table D-6. Intensity Distribution Values for All Subjects in the Situational Stress Experiment; Intensity Values are Expressed in dD re : lmv .

PAGE 164

APPENDIX E TABULATED VALUES FOR SFF DISTRIBUTIONS

PAGE 165

Table E-l. Speaking Fundamental Frequency (SFF) Distribution for the Male Subjects in the Laboratory Stress Experiment; Values of SFF Are Expressed in Semitone (ST) Intervals. SFF ST Hz Speaking Condition Normal Stress Difference 20

PAGE 166

140 Table E-2. Speaking Fundamental Frequency (SFF) Distribution for the Female Subjects in the Laboratory Stress Experiment; Values of SFF Are Expressed in Semitone (ST) Intervals. SFF Speaking Condition S T Hz Normal Stress Difference 2 7

PAGE 167

147 Table E-3. Speaking Fundamental Frequency (SFF) Distribution for All Subjects in the Laboratory Stress Experiment; Values of SFF Are Expressed in Semitone (ST) Intervals.

PAGE 168

140 Table E-4. Speaking Fundamental Frequency (SFF) Distribution for the Male Subjects in the Situational Stress Experiment; Values of SFF Are Expressed in Semitone (ST) Intervals.

PAGE 169

149 Table E-5. Speaking Fundamental Frequency (SFF) Distribution for the Female Subjects in the Situational Stress Experiment; Values of SFF Are Expressed in Semitone (ST) intervals.

PAGE 170

150 Table E-6. Speaking Fundamental Frequency (SFF) Distribution for All Subjects in the Situational Stress Experiment; values of SFF Are Expressed in Semitone (ST) intervals. SFF

PAGE 171

APPENDIX F TABULATED VALUES FOR TIME-ENERGY DISTRIBUTIONS (TED)

PAGE 172

Table F-l. Moan Values for the Time-Energy Distributions (TED) L"of the Laboratory Stress Experiment; Value: Are the Per Cent of Time for Each of Nine Energy Levels . Energy Level Speaking Condition MALES Normal 72.7 56.1 43.0 33.1 25.3 18.5 13.3 8.8 5.8 Stress 73.9 58.2 45.5 35.5 27.1 19.9 13.9 9.1 5.8 Difference 1.2 2.1 2.5 2.4 1.8 1.4 0.6 0.3 0.0 X d = 1.37, s d = 0.92, t d = 4.22, df = 8 (t_ Q5 = 2.306) FEMALES Normal 73.3 53.7 39.4 29.3 21.4 15.1 10.7 6.9 4.6 Stress 73.5 54.7 41.6 31.3 23.4 16.6 11.8 8.3 5.1 Difference 0.2 1.0 2.2 2.0 2.0 1.5 1.1 1.4 0.5 X d = 1.32, s, = 0.69, t, 5.42, df = 8 u d d (t_ Q5 = 2.306) OVERALL Normal 73.0 55.1 41.5 31.5 23.7 17.1 12.2 8.0 5.3 Stress 73.8 56.8 43.9 33.8 25.6 18.6 13.0 8.8 5.5 Difference 0.8 1.7 2.4 2.3 1.9 1.5 0.8 0.8 0.2 X d = 1.38, s d = 0.76, t d = 5.14, df = 8 (t >Q5 2.306) 152

PAGE 173

153 Table F-2 . Mean Values for the Time-Energy Distributions (TED) for the Situational Stress Experiment; Values Are the Per Cent of Time for Each of Nine Energy Levels. Speaking Condition Energy Level A. MALES Normal 74.3 58.3 45.8 35.6 26.8 19.4 13.6 9.4 5.8 Stress 79.4 62.3 48.9 38.5 30.1 23.1 17.5 12.8 9.4 Difference 5.1 4.0 3.1 2.9 3.3 3.7 3.9 3.4 3.6 X d = 3.67, s d = 0.65, df = 8, t d = 15.97 {t Q5 = 2.306) B. FEMALES Normal 73.9 52.3 38.3 27.7 20.1 13.9 9.3 6.5 4.4 Stress 78.1 58.3 45.6 36.3 27.8 20.9 15.4 11.2 7.9 Difference 4.2 6.0 7.3 8.6 7.7 7.0 6.1 4.7 3.5 X d = 6.12, s d = 1.71, df = 8, t d = 10.12 (t Q5 = 2.306) C . OVERALL Normal 74.1 55.8 42.7 32.3 24.0 17.1 1.1.8 8.2 5.2 Stress 78.9 60.7 47.5 37.6 29.2 22.2 16.6 12.1 8.8 Difference 4.8 4.9 4.8 5.3 5.2 5.1 4.8 3.9 3.6 X = 4.71, s d = 0.58, df =8, t d = 22.97 (t.05 = 2 -30 6 )

PAGE 174

REFERENCES Acker, M. and Reynolds, P. On the Assessment of Anxiety: III. by Self Rating, Psychol. Rep ., 19, 251-254, 1966. Almeida, A., Fleischmann, G., Ileike, G., and Thormann, E. Short Time Statistics of the Fundamental Tone in Verbal Utterances Under Psychic Stress, Proceeding of the Eighth International Congress of Phonetic Sciences , Leeds, England, August, 1975. Appley, M. K. and Trumbull, R. On the Concept of Psychological Stress, in Ps ychological Stress; Is sues in Research , Appley, M. H. and Trumbull, R. (eds . ) , Meredith Publishing Co., New York, 1967. Arnold, M. B. Stress and Emotion, in Psychological Stress; iss ues in Research , Appley, M. H. and Trumbull, R. (eds.), Meredith Publishing Co., New York, 1967. Dankart, C. P. and Elliot, R. Heart Rate and Skin Conductance in Anticipation of Shocks with Varying Probability of Occurrence, P sychophys iology , 11(2), 160-174, 1974. Barland, G. II. Use of Voice Changes in the Detection of Deception, 86th Meeting of the Acoustical Society of America , Los Angeles, California, October, 1973. Basowitz, II., Persky, H., Korchin, S. J. and Gr inter, R. R. Anxiety and Stress , McGraw-Hill, Inc., New York, 1955. Bennett, R. II., Jr. Hagoth; Fundamentals of Voice Stress An alysis , Hagoth Corp . , Issaquah, Washington, 1977. Blair, D. A., Glover, W. E., Greenfield, A. D. M. and Roddic, I. C. Excitation of Cholinergic Vasodilator Nerves to Human Skeletal Muscles During Emotional Stress, J. Physiol ., 148, 633-647, 1959. 154

PAGE 175

15' Blalock, H. M., Jr. Social Statistic s, McGraw-Hill, New York, N.Y., 1972. Bonner, R. Changes in the Speech Pattern Under Emotional Tension, Amer . J. Psychol . , 56, 262-273, 1943. Bowers, K. S. The Effects of UCS Temporal Uncertainty of Heart Rate and Pain, Psychophysiology , 25(3), 175-185, 1971. Brod, J., Fencil, V., Hejl, Z. and Jirka, J. Circulatory Changes Underlying Blood Pressure Elevation During Emotional Stress (Mental Arithmetic) in Normotensive and Hypertensive Subjects, Clinical Science , 18, 269279, 1959. Cohen, A. Estimating the Degree of Schizophrenic Pathology from Recorded Interview Samples, J. Clin. Psychol ., 17, 403-408, 1961. Darwin, C. The Expression of the Emotions in Man and Animals , Philosophical Library, New York, 1955 (Originally published in 1872) . Deane, G. E. Human Heart Rate Responses During Experimentally Induced Anxiety, J. Exper. Psychol ., 61, 489-493, 1961. Doherty, E. T. and Hollien, II. Multiple-Factor Speaker Identification of Normal and Distorted Speech, J. Phonetics , 6, 1-8, 1978. Fairbanks, G. and Hoaglin, L. W. An Experimental Study of the Durational Characteristics of the Voice During the Expression of Emotion, Speech Mono . , 8, 85-90, 1941. Fairbanks, G. and Provnost, W. An Experimental Study of the Pitch Characteristics of the Voice During the Expression of Emotion, Sp eech Mono . , 6, 87-104, 1939. Farr, J. L. and Seaver, W. B. Stress and Discomfort in psychological Research: Subject Perceptions of Experimental Procedures, Amer. Psychol., 30(7), 770-773, 1975.

PAGE 176

1136 Friedhoff, A. J., Alpert, M. and Kurtzberg, R. L. An Electro-Acoustical Analysis of the Effects of Stress on Voice, J. Neuropsych ., 5(5), 265-272, 1964. Hecker, M. H. L., Stevens, K. N., von Bismarck, G. and Williams, C. E. Manifestation of Task-Induced Stress in the Acoustic Speech Signal, J. Acoust. Soc. Amer ., 44, 993-1001, 1968. Hollien, H. and Majewski, W. Speaker Identification by Long-Term Spectra Under Normal and Distorted Speech Conditions, J. Acoust. Soc. Amer ., 62, 975-980, 1977. Hollien, H., Majewski, W. and Hollien, P. Perceptual Identification of Voices Under Normal, Stress, and Disguise Speaker Conditions, J. Acoust. Soc. Amer ., 56, S53, 1974. " ' "" " " " Hollien, H., Michel, J., and Doherty, E. T. A Method for Analyzing Vocal Jitter in Sustained Phonation, J. P honetics , 1, 85-91, 1973. Issihiki, N. Regulatory Mechanism of the Pitch and Volume of Voice, Oto-Rhino-Laryngo lo gy C linic , Kyoto, 52, 1065-1094, 1959. Itil, T. M., Simeon, J. and Coffin, C. Qualitative and Quantitative EEG in Psychotic Children, Pis. Nerv . Syst. , 37, 247-252, 1976. Johnson, C. C. Temporal Parameters Within the Speech Signal Applied to Speaker Identification, Unpublished Dissertation, University of Florida, 1978. Kubis, J. F. Comparison of Voice Analysis and Polygrath a s Lie Detection Procedures . Contract DAAD05-72-C-0217, Prepared for U. S. Army Land Warfare Laboratory, Aberdeen Proving Ground, Maryland, August, 1973. Kuroda, I., Fujiv^ara, O., Okamura, N. and Utsuki, N. Method for Determining Pilot Stress Through Analysis of Voice Communication, Avia. Space and Environ. Med ., 47 (5) , 528-533, 1976.

PAGE 177

157 Ladefoged, P. and McKinney, N. P. Loudness Sound Pressure, and Subglottal Pressure in Speech, J. Acoust. Soc. Amer ., 35, 454-460, 1963. Lazarus, R. S. Ps ychological S tress and the Coping Process , McGraw-Hill, Inc., New York, 1966. Levitt, E. E. The_ Psycholog y of Anxiety , The Bobbs-Merr ill Co., Inc., Indianapolis, 1967. Lieberman, P. Perturbations in Vocal Pitch, J. Acoust . Soc. Amer ., 33(5), 597-603, 1961. McGlone, R. E. Tests of the Psychological Stress Evaluator (PSE) as a Lie and Stress Detector, P roceedings of the 1975 Car nahan Conference on Crime Countermeasures , 83-86, May, 1975. Ortleb, R. An Objective Study of Emphasis in Oral Reading of Emotional and Unemotional Material, Speech Mono ., 4, 56-58, 1937. Os twa Id, P . F . Soundmaking: The Acous tic C omm unication of Emotion. Charles C. Thomas, Springfield, 111., 1963. Schramm, W. The Acoustical Nature of Accent in American Speech, A mer. Speech , 12, 49-56, 1937. Silverman, F. II. and Silverman, E. M. Effects of Threat of Shock for Being Disfluent on Fluency of Normal Speakers, Percept. Motor Skills , 41, 353-354, 1975. Simonov, P. V. and Frolov, M. V. Utilization of Human Voice for Estimation of Man's Emotional Stress and State of Attention, Aero. Med ., 44, 256-258, 1973. Simonov, P. V. and Frolov, M. V. Analysis of the Human Voice as a Method of Controlling Emotional State: Ach.i cvements and Goa ] s , Avia. Space Enviro n. Med. , 48(1) , 23-25, 1977. Simonov, P. V., Frolov, M. V. and Taubkin, V. L. Use of the Invariant Method of Speech Analysis to Discern the Emotional State of Announcers, Av ia_. Space Environ. Med., 46(8), 1014-10.16, 1975.

PAGE 178

158 Sovijarvi, A. How Does Our Voice Reflect the Habits and Routines of Our Work, Verbum Habet Sakala , Uppsala, 1974. Tiffin, J. and Steer, M. D. An Experimental Analysis of Emphasis, Speech Mono . , 4, 69-74, 1937. Wherry, R. J. Model for the Study of Psychological Stress Aero. Med ., 37, 495-500, 1966. Williams, C. E. and Stevens, K. N. On Determining the Emotional State of Pilots During Flight: An Exploratory Study, Aero. Med ., 40(12), 1369-1372, 1969. Williams, C. E. and Stevens, K. N. Emotions and Speech: Some Acoustic Correlates, J. Ac oust. Soc. Amer ., 52(4), 1238-1250, 1972. Zuckerman, M., Lubm, B., Vogel, L. and Valerius, E. Measurement of Experimentally Induced Affects, J. Consult. Psychol ., 28, 418-425, 1964.

PAGE 179

BIOGRAPHICAL SKETCH James Woodrow Hicks, Jr., was born September 24, 1950, in Jackson, Mississippi. At the age of five months he moved to the Panama Canal Zone. In June, 1968, he was graduated from Balboa High School. From September, 1968 until June, 1969 Mr. Hicks attended Canal Zone College. In September, 1969/ he enrolled at the University of Florida and in March, 1973, received the degree of Bachelor of Science in Electrical Engineering. From April, 1973, until September, 1973, Mr. Hicks was employed as an engineering technician with the Panama Canal Company. In September, 1973, he enrolled in the Graduate School of the University of Florida. From September, 1973, until June, 1974, he worked as a research assistant at the Communication Sciences Laboratory. From June, 1974* until September, 1977, Mr. Hicks held a Pre-doctoral fellowship at the Institute for Advanced Study of the Communication Processes (IASCP) at the University of Florida. In August, 1976, Mr. Hicks received the degree of Master of Arts. From January, 1978 , until the present Mr. Hicks has worked as a graduate 159

PAGE 180

160 research assistant for IASCP and the Department of Mechanical Engineering at the University of Florida. From August, 1976, until the present time he has pursued his work toward the degree of Doctor of Philosophy. Mr. Hicks ' specialty fields are experimental phonetics and underwater communications.

PAGE 181

I certify that I have read this study and that m my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. / -J^CaaXvx. Harry Hollien, Chairman Professor of Linguistics and Speech I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. \\ gMM :*iL/VA,^ &"V»— Howard B. Rothman Associate Professor of Speech I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the deqree of Doctor of Philosophy, William S. Brown, Jr. Associate Professor of Speech rtifv that I have read this study and that in my k rnold Paige Associate Professor Engineering fi Electrical

PAGE 182

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Kenneth J. Gerhardt Assistant Professor of Speech This dissertation was submitted to the Graduate Faculty of the Department of Speech in the College of Liberal Arts and Sciences and to the Graduate Council, and was accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy. December, 1979 3^ i -' /'
PAGE 183

UNIVERSITY OF FLORIDA 3 1262 08552 9807