An analysis of the durational aspects of connected speech with reference to stuttering.

MISSING IMAGE

Material Information

Title:
An analysis of the durational aspects of connected speech with reference to stuttering.
Alternate title:
Durational aspects of connected speech with reference to stuttering
Physical Description:
viii, 109 leaves. : ill. ; 28 cm.
Language:
English
Creator:
Flanagan, Bruce Clement, 1929-
Publication Date:

Subjects

Subjects / Keywords:
Stuttering   ( lcsh )
Speech   ( lcsh )
Speech thesis Ph. D   ( lcsh )
Dissertations, Academic -- Speech -- UF   ( lcsh )
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Thesis:
Thesis--University of Florida.
Bibliography:
Bibliography: leaves 104-107.
Additional Physical Form:
Also available online.
General Note:
Manuscript copy.
General Note:
Vita.
Statement of Responsibility:
by Flanagan, Bruce Clement

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
oclc - 18312666
System ID:
AA00024004:00001

Table of Contents
    Title Page
        Page i
        Page i-a
    Acknowledgement
        Page ii
    Table of Contents
        Page iii
    List of Tables
        Page iv
        Page v
        Page vi
    List of Figures
        Page vii
        Page viii
    Chapter 1. Introduction
        Page 1
        Page 2
        Page 3
        Page 4
        Page 5
        Page 6
        Page 7
        Page 8
        Page 9
        Page 10
        Page 11
        Page 12
        Page 13
        Page 14
        Page 15
    Chapter 2. Method
        Page 16
        Page 17
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
        Page 23
    Chapter 3. Results
        Page 24
        Page 25
        Page 26
        Page 27
        Page 28
        Page 29
        Page 30
        Page 31
        Page 32
        Page 33
        Page 34
        Page 35
    Chapter 4. Discussion
        Page 36
        Page 37
        Page 38
        Page 39
        Page 40
        Page 41
        Page 42
        Page 43
        Page 44
    Chapter 5. Summary
        Page 45
        Page 46
        Page 47
    Appendices
        Page 48
    Appendix A. Preliminary research: Durational properties of silence intervals during oral reading
        Page 49
        Page 50
        Page 51
        Page 52
        Page 53
        Page 54
        Page 55
        Page 56
        Page 57
        Page 58
        Page 59
        Page 60
        Page 61
        Page 62
        Page 63
        Page 64
        Page 65
        Page 66
        Page 67
        Page 68
        Page 69
        Page 70
        Page 71
        Page 72
        Page 73
        Page 74
        Page 75
        Page 76
        Page 77
        Page 78
        Page 79
    Appendix B. Supplementary statistical summaries
        Page 80
        Page 81
        Page 82
        Page 83
        Page 84
        Page 85
        Page 86
        Page 87
        Page 88
        Page 89
        Page 90
    Appendix C. Z score data for the individual stuttering speakers
        Page 91
        Page 92
        Page 93
        Page 94
        Page 95
        Page 96
        Page 97
        Page 98
        Page 99
        Page 100
        Page 101
        Page 102
        Page 103
    References
        Page 104
        Page 105
        Page 106
        Page 107
    Biographical sketch
        Page 108
        Page 109
        Page 110
        Page 111
Full Text










AN ANALYSIS OF THE DURATIONAL ASPECTS OF CONNECTED SPEECH WITH REFERENCE TO STUTTERING












By

BRUCE C. FLANAGAN














A DISSERTATION PRESENTnD TO T14E GRADUATE COUNCIL
THE UNIVERSITY OF FLORIDA
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE
DEGREE OF DOCTOR OF PHILOSOPHY














UNIVERSITY OF FLORIDA June, 1966




































































CA

ICY













ACKNOWLEDGEMENTS


The author wishes to express his gratitude to Doctor M~c~enzie W. Buck for his continuing interest and encouragement throughout their academic association. Grateful acknowledgment is expressed to Doctors G. Paul Noore and E. Porter Horne for their guidance during the author's doctoral program and preparation of this dissertation. Sincere appreciation is expressed to Doctor Wilse B. Webb for his salient and stimulating contributions to the author's doctoral program.

Special acknowledgment is given to Doctors Thomas B. Abbott and Kenneth R. Bzoch for their cooperation in making subjects and facilities available for obtaining data. Doctor Leslie F. Vlalpass and Dean Edwin P. Martin of the University of South Florida are especially acknowledged for their continuing support and for the utilization of facilities to collect the data for the major portion of this dissertation.

The Operational Applications Laboratory, Air Force Cambridge Research Center is acknowledged for funding the collection of the data for the preliminary research; Contract No. AFl9(604)-6127, Israel Goldiamond, Principal investigator.


ii













TABLE OF CONTENTS

Page

ACKNOWLEDGMENTS ii

LIST OF TABLES iv

LIST OF FIGURES vii

Chapter

I. INTRODUCTION 1
Review of the Literature 4

II. METHOD 16
Selection of Subjects 16
Apparatus 17
Recording Procedures 20
Oral Reading Time and Disfluency Measures 22

III. RESULTS 24
Reliability of Individual and Group Measures 24
Comparison of Fluent and Difluent Speech
Sanp! es 30
Effects of Time of Day and ExperimenterSubAject Combinations 34

IV. DISCUSSION 36
Implications for Application and Further
Research 41

V. SUMMARY 45

APPENDICES 48

A. PRELIMINARY RESEARCH: DURATIONAL PROPERTIES
OF SILENCE INTERVALS DURING ORAL READING 49

B. SUPPLEMENTARY STATISTICAL SUMM'ARIES 80

C. Z SCORE DATA FOR THE INDIVIDUAL STUTTERING
SPEAKERS 91

REFERENCES 104

BIO GRAPHICAL SKETCH 108

iii











LIST OF TABLES

Table Page

1. Means, standard deviations and correlations of Trials I and 1I for frequency of speech events
of 64 fluent speakers . . . . . . 26

2. Means, standard deviations and correlations of
Trials I and II for frequency of silence events
of 64 fluent speakers . .. .. .. . .. 29

3. Means, standard deviations and t tests
comparing 12 stutterers with 16 fluent speakers
for frequency of speech events . . . . 32

4b. Means, standard deviations and t tests
comparing 12 stutterers with 16 fluent speakers
for frequency of silence events . . . . 33

5. Number of silence intervals recorded at each
duration, when same tape was run four times .. 56
6. Oral reading rate for normal subjects
expressed in words per minute . . . . 57

7.Percent of disfluent words for normal subjects. 60

8. Oral reading rate for stuttering subjects expressed in words per minute . . . . 62

9. Percent of disfluent words for stuttering
subjects . . . . . . . . . 64

10. Reliability coefficients of the Multiple-Class
Time Analyzer for speech and silence events .. 81

11. Reliability of reading time and disfluency
measurement procedures . . . . . . 82

12. Summary of analysis of variance for experimenter-subject combinations by time of day
variables for reading time of Trial I . . 85

13. Summary of analysis of variance for experimenter-subject combinations by time of day
variables for reading time of Trial II . . 83

iv










Table Page

14. Summary of analysis of variance for experimenter-subject combinations by time of
day variables for disfluencies occurring
during Trial I . . . . 0 . . 84

15. Summary of analysis of variance for experimenter-subject combinations by time of
day variables for disfluencies occurring
during Trial II . . . . . .. 84

16. Summary of analysis of variance for
experimenter-subject combinations by time of day variables for frequency of silence events
between 100-149 milliseconds on Trial I. . 85

17. Summary of analysis of variance for experimenter-subject combinations by time of day variables for frequency of speech events
between 150-199 milliseconds on Trial I. . 85

18. Summary of analysis of variance for experimenter-subject combinations by time of day variables for frequency of speech events
between 400-499 milliseconds on Trial I. . 86

19. Summary of analysis of variance for experimenter-subject combinations by time of day variables for frequency of silence events
between 450-499 milliseconds on Trial I. . 86

20. Summary of analysis of variance for experimenter-subject combinations by time of day variables for frequency of silence events
between 500-549 milliseconds on Trial I .. 87

21. Summary of analysis of variance for experimenter-subject combinations by time of
day variables for the frequency of silence
events between 600-649 milliseconds on Trial I 87

22. Summary of analysis of variance for experimenter-subject combinations by time of day variables for frequency of speech events
between 650-699 milliseconds on Trial I .. 88



v










Table Page

23. Summary of analysis of variance for experimenter-subject combinations by time of day variables for frequency of speech events
between 250-299 milliseconds on Trial 11 88

24. Summary of analysis of variance for experimenter-subject combinations by time of day variables for frequency of speech events
between 350-399 milliseconds on Trial II . 89

25. Summary of analysis of variance f or experimenter-subject combinations by time of
day variables for frequency of silence events
between 650-699 milliseconds on Trial II . 89

26. Summary of analysis of variance for experimenter-subject combinations by time of day variables for frequency of speech events
between 700-749 milliseconds on Trial II . 90

27. Summary of analysis of variance for experimenter-subject combinations by time of day variables for frequency of silence events
beyond 750 milliseconds on Trial II . . . 90






















vi














LIST OF FIGURES

Figure Page

1. A composite plot of the cumulative frequency
of silence intervals for five fluent subjects
reading a 1024 word passage six times . . 67

2. The cumulative frequency of silence events for
Subject 1 in Trials 1, 111, and V . . . 70

3. The cumulative frequency of silence events for
Subject 2 on Trials I, III, and V . . . 71

4. The cumulative frequency of silence events for
Subject 3 on Trials I, III, and V. & 0 0 73

5. The cumulative frequency of silence events for
Subject 4 on Trials 1, 111, and V. 74

6. The cumulative frequency of silence events for
Subject 5 on Trials I, III, and V. 0 * 0 0 76

7. The cumulative frequency of silence events for
Subject 6 on Trials I, III, and V . . . 77

8. Z scores at significant speech and silence
class intervals for male stutterer, age 22 92

9. Z scores at significant speech and silence
class intervals for female stutterer, age 18 93

10. Z scores at significant speech and silence
class intervals for male stutterer, age 23 94

11. Z scores at significant speech and silence
class intervals for female stutterer, age 18 95

12. Z scores at significant speech and silence
class intervals for male stutterer, age 22 96

13. Z scores for significant speech and silence
class intervals for male stutterer, age 20 9?

14. Z scores for significant speech and silence
class intervals for male stutterer, age 28 98 vii











Figure Page

15. Z scores for significant speech and silence
class intervals for male stutterer, age 19 . 99

16. Z scores for significant speech and silence
class intervals for male stutterer, age 29 . 100

17. Z scores for significant speech and silence
class intervals for male stutterer, age 21 . 101

18. Z scores for significant speech and silence
class intervals for male stutterer, age 27 . 102

19. Z scores at significant speech and silence
class intervals for male stutterer, age 21 . 103













CHAPTER I

INTRODUCTION


The purposes of this research were to establish

variance estimates for frequency distributions of speech and silence events for fluent speakers during conditions of oral reading, to determine the reliability of these dimensions of vocal behavior on a test-retest basis, and to assay the effects of time of day and experimenter variables on these measures, Contingent upon the reliability of the variance estimates, a comparison of the frequency distributions was made between stutterers and fluent speakers for speech and silence events.

Although the disfluency phenomenon has been an area of continuing interest to scientists and clinicians, recently, Carroll (1964) indicated surprise at the lack of definitive findings concerning disfluency in view of the considerable quantity of research devoted to the topic. This lack of definitive results is a function of the measurement procedures employed. Research devised to assess the effects of certain independent variables on disfluency have relied on human observers to count or rate this phenomenon. Reliance on an observer definition of disfluency may be


1








2

considered a special case of psychophysical analysis where the properties of the stimuli are unknown. Current psychophysical procedures admit varying degrees of observer bias (Goldiamond, 1958) which could obviate the usefulness of observer defined procedures for a detailed analysis of disfluency.

The information accumulated from previous studies has demonstrated some consistent findings relative to the properties of disfluency. Disfluent speakers (stutterers) as a group have been found to use significantly more time in reading a given passage than do fluent speakers (Johnson, 1961). Also, within stutterers as a group, there has been reported a high positive correlation between the judged frequency of disfluency and reading time (Sander, 1961). From these observations it appears tenable to assume observer defined disfluencies occurring during connected speech possess correlates which occupy time.

The problem then arises in discriminating between the fluent and disfluent durational characteristics of speech. Speech and silence events occurring during connected speech can be separated and tabulated by their durations. These data may then be assigned to arbitrary class intervals which permit construction of frequency distributions for speech and silence events. It was hypothesized: Under specified controlled conditions, the frequency distributions of speech











and silence events will demonstrate reliable mean estimates at the selected class intervals. Contingent upon the accuracy of this hypothesis, it was further hypothesized: The frequency distributions of speech and silence events for a group of disfluent speakers will deviate from the fluent speakers' means at one or more class intervals. This hypothesis was predicated on the cited observation that disfluent speakers use significantly more time to read a given passage.

In addition to recognizing capricious variability as a potential source of limitation, unavoidable variations within the controlled experimental conditions would also seriously impair the usefulness of these data. Time of day and sex of experimenter in combination with sex of the subject are variables within the controlled condition which cannot be avoided efficiently. Therefore, two additional hypotheses were advanced: (1) The time of day during which subjects are run does not significantly effect frequency distributions of speech or silence events; (2) The sex of the experimenter in combination with sex of the subject does not significantly effect frequency distributions of speech or silence events.











Review of the Literature

Stuttering Literature. Literature pertaining to the measurement of disfluency can be broadly categorized by observer defined procedures, or by correlates of observer defined procedures. The disfluency correlates can be categorized by the measurement procedures employed. The relevant literature will be reviewed in relation to acoustical, physiological, rate, and durational distribution measurement procedures. Discrimination of disfluent speech cannot be separated from the operational procedure used in measuring it. To avoid biased assumptions regarding the nature of the disfluency phenomenon, the pertaining literature will be discussed on an operational level.

Johnson (1961) reported a procedure, the Iowa Speech Disfluency Test, in which tape recordings were obtained under several conditions of speaking and oral reading. The observer listened to the recordings and categorized judged disfluencies into eight types.. This test procedure allowed as many replayings as necessary to satisfy accurate identification of the disfluencies. The eight categories were (1) interjections, (2) part-word repetitions, (3) word repetitions, (4) phrase repetitions, (5) revisions, (6) incomplete phrases, (7) broken words, and (8) prolonged sounds. Rate of speaking and oral reading were also computed, These data were then compared to norms for normal speakers and for








5

stutterers. Sander (1961) investigated the reliability of the Iowa Speech Disfluency Test. The reliability coefficients for total disfluencies and speaking rate were approximately .90 for a sample of 40 stutterers and high intercorrelations among subtests indicated redundancy. Young (1961) reported a similar technique for the evaluation of stuttering with the number of categories of disfluency types shortened to five for reasons of infrequent occurrence of some categories and confusion among others. These categories of disfluency were (1) interjections, (2) partword repetitions, (3) word-phrase repetitions, (4) prolongations, and (5) revisions. Minifie and Cooker (1964) suggested a disfluency index considering syllables uttered divided by the reading rate for a set passage. Data were presented suggesting reliability of this procedure for discriminating fluent from disfluent speakers. The techniques cited involved recording speech samples on tape under several conditions and a subsequent detailed analysis. Several replayings of each recorded speech sample were compared to typescripts and the judged disfluencies were categorized on two or more dimensions. Reliability coefficients on a test-retest basis (above .90) were obtained for total disfluency count and rate of speech.

An alternate procedure instructed an observer to

rate a speech sample on an equal appearing scale of stutter-








6

ing severity (Lewis and Sherman, 1951). An investigation comparing several rating scales and rater instructions (Cullinan, Prather, and Williams, 1963) estimated that four or more judges were necessary to obtain inter-judge reliability coefficients above .90 for speech samples of 20 seconds' duration. There is also a question as to whether or not rating periods of stuttering severity following these. procedures is adequate for a prediction of stuttering frequency. Sherman and Trotter (1956) reported correlations of approximately .60 between frequency and severity of individual moments of stuttering. They recommended that measures of frequency and severity of disfluency are both needed to define the speaker's disfluency.

These data are cited to emphasize that currently available observer defined techniques of estimating disfluency are time consuming to the clinician or researcher. Less tedious alternatives involve reliability risks. An authority (Milisen, 1957) suggested these measures need not be highly accurate, and stressed disfluent behavior as being highly variable from period to period or from situation to situation. Recent research does not report such variability unless experimental contingencies were manipulated (Goldiamond, 1965). Perhaps observer techniques of measuring disfluency will of necessity be time consuming or vague until basic psychophysics advances substantially. Implicit in









7

this statement is the assumption that correlates of observer defined disfluency exist.

Hill (194,a,b) reported comprehensive reviews of bio-chemical and physiological correlates of stuttering. For these two broad categories, his conclusions were essentially negative with respect to the existence of causally linked events. Subsequent research in these areas has been scant. Williams (1953) compared the muscle action potentials in stuttered and non-stuttered speech by electromyographic procedures. He concluded that although the EMG pattern certainly reflected preceding motor neuron activity, he did not find sufficient evidence to imply stutterers and non-stutterers differ neuro-physiologically from one another.

The acoustical differences between the speech of stutterers and normal speakers were reported by Travis (1927a). Using a phonophotographic technique, the speech of disfluent and fluent speakers was compared under conditions of 'non emotional propositional' speaking. It was reported that disfluent speakers exhibited (1) marked prolongations of 'tones,' (2) fluctuations of breath pressure preceding and following phonation, and (3) short vibrations before onset of voice wave. Certain disfluent speakers showed 'bizarre' waves in phonation curves which varied markedly in length. Groups of oscillations of high frequency but low amplitude and long series of oscillations








8

at approximately 500 cycles per second were also reported. None of these findings were commonly found in the phonophotographs of the fluent speakers.

A subsequent report (Travis, 1927b) compared the

voices of stutterers and normal speakers under conditions of emotive and non-emotive producing stimuli. The stutterers had more pitch variations than did the non-stutterers. After being subjected to emotion producing stimuli, the stutterers had less variability while the non-stutterers had more as compared to their respective baseline conditions.

Adams (1955) designed a study to determine the

existence of differences between pitch characteristics of stutterers and non-stutterers during conversation. Syllables were selected from speech samples obtained from groups of stutterers and fluent speakers. Recordings of these speech samples were submitted to phonophotographic analysis. Significantly smaller mean pitch inflections were observed for the stuttering group. Subjects from the designated groups also recorded the sustaine. vowel (i), as in me2et, at each mean pitch level. Stutterers exhibited limited variation in pitch and the more severe stutterers showed the fewest variations. These studies have been concerned with comparing acoustical measures of speech and, sustained tones between groups of stutterers and normally fluent speakers.








9

Bryngleson (1932) reported acoustical phonophotographic analysis of vocal disturbances occurring during moments of stuttering for 17 adult stutterers. Phonophotographic records of the stutterer's voice during stuttering were reported to show (1) marked variation in the form, length, and intensity of consecutive waves; (2) marked reduction in tone variability; (3) a variety of isolated waves; (4) extreme variability in initial vocalization; and

(5) abnormal endings to vocalization. These observations were made relative to findings reported in studies of the phonophotographic analysis of normal speech.

The method for determining the moments of disfluency was not reported. If these citations represent a nearly complete survey of the literature on acoustical correlates of observer defined disfluency, decisions as to the relative usefulness of acoustical correlates rest with future research.

Bloodstein (1944) investigated the relationship between oral reading rate and stuttering severity. In these findings strong relationships between reading rate and disfluency and between reading rate and duration of disfluency were suggested. The reported relationships apparently held for the stutterers during moments of fluency. Similar findings have been reported by Roberts (1950), Robinson (1951) and Goldiamond (1965). Sander (1961)









10

reported a product moment correlation of .86 between disfluency and speaking rate.

The variability of frequency distributions of speech and silence events may pose a question. Will a stutterer with a judged high frequency of interjections have a different frequency distribution pattern from that of fluent speakers? Travis (192?a) reported the frequency distribution of speech events comparing a fluent speaker and a severe stutterer. Class intervals of 500 milliseconds and samples of conversational speech were used for the study. The stutterer exhibited fewer short speech events and substantially more frequent long speech events.

A paper presented at the 1958 American Speech and

Hearing Association Convention (Roe and Derbyshire) reported systematic differences between stutterers and normal speakers for durational distributions of silence intervals during conversational speech. In it, a class interval size of 50 milliseconds was utilized. Data were obtained by hand measuring photographs of cathode ray oscilloscope speech displays. The stutterers' conversational speech showed fewer intervals of short durations of silence and more frequent intervals of longer durations as compared to fluent speakers.

Preliminary Research. The report by Travis and the paper by Roe and Derbyshire represent the extent of current









11

information concerning the systematic relationship between durational correlates and observer defined disfluencies. Noting this lack of information,thsuhodiprlmny research designed to replicate the differences reported by Roe and Derbyshire. This preliminary investigation, utilizing an automatic analysis device, obtained further information concerning the frequency distributions of silence events which occurred during fluent and disfluent oral reading and related these data to disfluency frequency and reading time,

Tape recordings of five adult fluent speakers and six stutterers were made. The chosen reading passage was read consecutively six times by each subject. Subjects were selected to provide a range of reading and disfluency rates. The tape recordings were played to an automatic device capable of classifying silence events into thirty-three class intervals with the lowest observable silence event being .05 seconds. The device cumulatively recorded the number of silence events by length of duration. An observer listened to the recordings and tabulated the number of disfluencies. Reading time per passage was also obtained. The frequency distributions for the fluent subjects were plotted on a composite graph. The frequency distributions of silence events for the disfluent subjects were plotted individually, comparing each subject to himself on the several trials and to the composite graph of the fluent speakers.








12

The curves for cumulative frequency distributions of silence events of fluent subjects with the fastest reading times were approximately the shape of an inverted L. As oral reading time decreased, the point of inflection on the curve was more gradually approached, with the curves of the slower reading subjects exhibiting a greater absolute frequency of silence intervals. With these subjects, shifts in frequency distributions from trial to trial were not observed.

The frequency distribution of silence events for the disfluent speakers differed quantitatively from trial to trial, between subjects, and from the composite curves of the fluent speakers. Disfluent speakers had fewer silence events below 120 milliseconds than did the fluent speakers. There was a tendency for the curves associated with the highest disfluency percentage to be the most dissimilar. These had characteristics of increased acceleration followed by decreased acceleration. As the disfluency percentage more closely approximated that of the fluent speakers, the curves of frequency distributions of silence events more nearly approximated the fluent subject's data. The cumulative frequency distributions of a disfluent subject, whose disfluency rate was within normal range, were comparable to the fluent speaker's composite curve for silence intervals. A detailed report of this preliminary research is included in Appendix A.








13

Frequency Distribution of Fluent Speech. In recent years a number of investigators have reported instruments for the automatic analysis of speech durations as they occur in connected speech samples (Hargreaves and Starkweather, 1959 and Verzeano, 1950). These devices used voice-operated electronic circuits which detected and counted total number of speech intervals in a speech sample falling within a set range of durations. These have been accompanied by reports which justify ignoring silence events below .5 seconds (Hargreaves, 1960 and Verzeano, 1951) as not being relevant to their experimental topics. The topics concerned discrimination among speech behavior in various environmental

settings and tasks.

Minifie (1963) reported a device capable of detecting speech or silence events in class intervals as short as ten milliseconds. A limitation of the device was the necessity of running each tape 52 times to obtain a cumulative distribution of speech and silence events for the speech samples contained. Using this analysis device, Minifie attempted to assay the effects of (1) instructions with respect to reading rate and vocal effort, (2) mean word length, and

(3) oral reading versus impromptu speaking. Speech samples of twelve graduate students were used. Changes in oral reading rate werelargely accounted for by fewer silence events of long durations such as would occur between phrases








14

and sentences. Instructions to increase vocal effort correlated with detectable increases in the frequency of speech events associated with the duration of vowels. The converse was observed when the speakers were instructed to decrease vocal effort. Negotiable effects on frequency distributions for speech and silence events were reported for three different reading passages which varied by mean word length measured in syllables. The effects of impromptu speaking versus baseline oral reading were observed by the frequency distribution of fewer silence events of short duration (125 milliseconds to 500 milliseconds), and by an increase in the longer durations of silence.

It should be emphasized that the shape of the cumulative durational frequency curves for silence events approximated those observed in the cited preliminary research for fluent subjects. In both the preliminary and Minifie's research, faster oral reading rates were associated with a sharp point of inflection and slower reading rates with a more gradual inflection point and greater frequencies of periods of silence,

Generalization from these data is hazardous because

cumulative durational distributions do not provide a straightforward and readily interpretable method of dealing with subject to subject variability. Each class interval is not only dependent on the count which occurs within a fixed time








15

range but also on events which occur in every class interval temporally before it. These unique plots were necessarily employed in the above studies because the automatic apparatus used in each investigation limited the time duration of each class interval to unequal periods. These unequal periods seriously limited graphic and/or numerical presentation methods and analysis procedures, Subsequent research was proposed using instrumentation which would permit collection of data in class intervals of equal temporal values, thus allowing estimates among and within subject variability at given points along the durational parameters of vocal behavior.













CHAPTER II

METHOD


The procedures followed in this research may be

described in reference to selection of subjects, apparatus, recording procedures and measures of oral reading time and disfluency.

Selection of Subjects

Thirty-two male and thirty-two female young adults served as fluent subjects. They were obtained from a Behavioral Science course at the University of South Florida. Their ages ranged from 17 to 30 years. Any student who reported possessing a speech problem or demonstrated a speech problem to the experimenter was rejected as a subject. The young adult college population was selected for several reasons apart from availability. College age fluent speakers and stutterers compare'favorably with respect to language ability (Steer, 1936). Compared to the young adult population as a whole, subjects selected on the basis of probable academic success could be expected to demonstrate relatively greater homogeneity in respect to many measures

of vocal behavior.

Twelve stutterers, two female and ten male, served as a disfluent speaking group. Their ages ranged from 18 16








17

to 29 years. The disfluent subjects were obtained from the University of Florida's Speech and Hearing Clinics and from the Developmental Center at the University of South Florida. All reported having a communication problem called stuttering and had received a minimum of several monthsltreatment. Apj2aratus

The frequency distributions for speech and silence events of the oral readings were obtained by playing tape recorded speech samples to a voice-operated relay which converted speech or silence to DC pulses. The pulses were fed to an electronic multiple-class time analyzer with fifteen class intervals, the last class interval being without an upper limit.

The voice-operated relay and multiple-class time analyzer were treated to eliminate speech and silence events of durations less than 50 milliseconds. This decision was prompted by Minifie (1963) who reported a notch or gap in the distributions at the 40-50 milliseconds time interval which caused him to conclude that events below 4050 milliseconds were not likely to change with speech behavior. He also reported that speech and silence events shorter than 40 milliseconds contributed less than 3 per cent of the total speaking time. A practical argument for eliminating these durations was in the limitation of the apparatus which did not function reliably below 50 milli-








18

seconds. The time analyzer classified speech or silence events into fourteen 50 millisecond class intervals. Durations above 750 milliseconds were recorded in a fifteenth counter. Curves for the cumulative durational distributions reported in the preliminary research by the author and by Iiinifie, became asymptomic beyond 600 milliseconds. Based on these data, the durations between 50 and 750 milliseconds appeared to be the most productive portions of the frequency distributions for speech and silence events.

The gain control of the voice-operated relay was set to operate at 100 millivolts for white noise and square waves of 250, 500, and 1000 cycles per second. The noise level of each tape played to the voice-operated relay was set within five millivolts of the voltage required to operate the relay for these signals. These measurements were made with a Hewlett Packard 400B voltmeter.

Tape recordings of each speech sample were played to the durational analysis apparatus twice. Once to obtain the frequency distributions for speech events, and once to obtain the frequency distributions of silence events.

Calibration of the durational analysis device was

accomplished by presenting known time values of white noise to its input, The time values of these signals ranged from 50 to 750 milliseconds. The internal clock of the Hewlett Packard 521A Electronic Counter, which is capable of









19

clocking events in 0.1 millisecond units, was used for the time reference. Repeated checks of each class interval indicated the apparatus was functioning within three milliseconds tolerance.

Reliability of the durational analysis device was established by selecting at random sixteen taped speech samples and independently analyzing the results of each sample twice. Product-moment correlations were run and these results are reported in Appendix B, Table 10. Reliability coefficients for speech events ranged from .917 at 500-549 milliseconds to .992 at 750 milliseconds and beyond. The median correlation was .955 at 150-199 milliseconds. For silence events, the reliability coefficients ranged from .844 at 4-i50-499 milliseconds to .995 at 50-99 milliseconds with the median correlation being .957 at 200249 milliseconds. The durational analysis device used for this research can be commercially obtained from the G-rason Stadler Company, West Concord, Mlassachusetts as (1) E7300A Voice-Operated Relay, (2) E3950A Multiple-Class Time Analyzer, and (3) EllOOD Power Supply.

The tape recordings made at the University of South Florida and at the University of Florida were made in comparable Industrial Acoustics Company sound-treated rooms. All speech samples were recorded on a Wollensak T1500 tape recorder. The speed of the tape deck used was empirically








20

determined to be constant between record and play-back conditions within 2 per cent. RecordinE Procedures

Each subject was seated at a table in the soundtreated room with a microphone positioned approximately 18 inches from his mouth. The experimenter, seated across from the subject, asked identifying information such as name, age, level of education and marital status. The subject was then asked why he chose to attend college, what previous experience he had in recording his speech, and if he had a history of any speech problem. The purpose of this questioning was to accustom the subject to the experimental situation. The subject was then handed a copy of the 'Arthur' passage (Fairbanks, 1940) and told to read it orally as he ordinarily would. He was also cautioned to remain a constant distance from the microphone. These instructions followed those reported for the Iowa Speech Disfluency Test. During this reading, the volume control on the tape recorder was adjusted to the subject's speech intensity, and this also provided an opportunity for further adaptation to the speaking situa tion. Upon completion, a ten-second period of no oral activity in the recording room followed to allow for adjustment of the noise level for the subsequent durational analysis. The subject was then handed








21

a copy of' the 'average' reading passage constructed by Darley (1940O) and instructed to read it as he ordinarily would. He was again cautioned to remain a constant distance from the microphone. Upon completion of this reading, the fluent subjects made an appointment for the second recording session (Trial II).

Two possible sources of' variance which are routinely encountered in research involving significant sample size are the time of' day at which the sessions are run and the

-sex of' the experimenter in combination with the sex of' the subject. Either or both factors could conceivably possess discriminative instruction value to the subject with respect to response rate or intensity (Staats and Staats, 1963). Time of' day, in addition to conditioned excitory or inhibitory factors, may relate indirectly to fatigue factors (Pavlov, 1927) which also may alter the response topology. The decision to vary the sex of' the experimenters was based on cul Itural emphasis (Berelson and Steiner, 1964). Therefore, a male and a female Speech Pathologist were used as experimenters. Time of day was divided into four periods:

(1) 8am-l0am, (2) l0am-l2pm, (3) lpm-3pm, and (4) 3pm-5Pm. The number of subjects run in each condition was as follows:








22

Male Experimenter Female Experimenter

Male Subject Female Subject Mvale Subject Female Subj ect

8am- 4 4 4I 4

l0am- 4I 4 4 4

1pm- 4 4 4 4

3pm- 4 4 4 4

Oral Reading Time and Disfluency Mleasures

.Oral reading time was obtained by measuring the

length of' the recorded reading passage with a stop watch. Reading time was rounded to the closest 0.5 second. Disfluencies were marked by the experimenter on a copy of the passage as he listened to the recorded speech sample. For purposes of obtaining the total count, a disfluency was defined as an interjection or repetition of a sound, syllable, word or phrase. The repetition of a given word was counted as a single disfluency regardless of the number of times the word or portion of' the word was repeated, Similarly, the interjection of a single word or of an entire phrase was counted as a single disfluency. This procedure was selected because recent research (Sander, 1961, and Siegel and M~artin, 1965) reported that it yields high reliability coefficients (above .90) between and within experimenters. Between and within experimenter reliability coefficients were obtained for twelve and sixteen speech samples respectively.









23

Reliability correlations were obtained for both oral reading time and total disfluency frequency. These results are presented in Appendix B, Table 11. The between and within experimenter reliability coefficients were observed to be above .90 for reading time. The reliability coefficients for frequency of disfluency were observed to be .97 between experimenters and .79 within experimenters. Tuthill (1916) reported reliability coefficients of .72 as typical test-retest values for frequency of disfluency.













CHAPTER III

RESULTS


The results of this research are presented in reference to reliability of individual and group measures; comparison of fluent and disfluent speech samples; and effects of time of day and experimenter subject combinations. Each section will contain the information relevant to oral reading time and disfluency frequency, frequency distributions of speech events, and frequency distributions of silence events.

Reliability of Individual and Group Measures

A purpose of this research was to obtain frequency distributions of speech events and silence events that occurred during oral reading for fluent speakers and to test the stability of these means on a test-retest basis, Means, standard deviations and correlations for Trials I and II of fluent speakers were obtained for these measures. The reliability of the data was tested on an individual (correlation) and group (t test) basis.

Oral ReadinE Time and Disfluency Frequency. The mean reading time for fluent speakers on Trial I was 95-01 seconds with a standard deviation of 9.1?. For Trial II, the mean 24









25

reading time was 92.41 seconds and the standard deviation was 7.96. The correlation between Trials I and II for reading time was .945. For the mean difference between trials, the t ratio was 6.67 which is significant beyond the .01 level of confidence.

The mean of the total disfluency frequency count for these 64 fluent speakers on Trial I was 5.59 with a standard deviation of 2.51, and on Trial 11 3.11 with a standard deviation of 2.55. The correlation between Trials I and II for disfluency frequency was .607. For mean difference between trials, the t ratio was 1.78, which failed to reach significance at the .05 level of confidence.

Frequency Distributions of Speech Events. For the purposes of this research, speech events were arbitrarily defined as vocalizations which transcended the noise level of the recording apparatus and exceeded a duration of 50 milliseconds. Table 1 presents the means, standard deviations and correlations between Trial I and Trial II for speech events of fluent speakers. The mean frequency of speech events between 50-99 milliseconds for Trial I was 64.25 with a standard deviation of 15.56, and for Trial II 64.81 with a standard deviation of 15.09, The mean frequency of speech events between 100-149 milliseconds for Trial I was 52.50 with a standard deviation of 9.81, and for Trial II 50.55 with a standard deviation of 9.12. In the third class








26

Table 1. Means, standard deviations and correlations of
Trials I and II for frequency of speech events of 64
fluent speakers.

Time in Mean Standard Correlation
milli- Deviation
seconds
Trial I Trial II Trial I Trial II r I II
50- 99 64.25 64.81 13.56 13.09 .713
100-149 52.30 50.33 9.81 9.12 .246
150-199 44.00 44.13 5.97 7.38 .150
200-249 29.34 29.83 6.02 6.12 .410
250-299 19.22 19.42 4.57 4.37 .123
300-349 15.55 15.70 4.82 4.45 .163
350-399 11.13 11.89 3.47 4.55 .311
400-449 7.75 7.72 3.61 4.04 .366
450-499 4.59 4.28 2.55 2.49 .279
500-549 2.98 2.97 1.97 2.10 .375
550-599 2.08 1.83 1.69 1.56 .270
600-649 1.63 1.36 1.69 1.34 .307
650-699 1.03 1.20 1.28 1.21 .241
700-749 .72 .73 .92 .91 .232
750-X 1.69 1.30 3.62 1.50 .134









2?

interval, 150-199 milliseconds, the mean frequency of speech events was 44.00 for Trial I and 44.13 for Trial II. The standard deviations for these trials were 5.97 and 7.38 respectively. This decrease in mean frequency of speech events continued through the twelfth class interval, 600649 milliseconds, where the mean frequency was 1.63 with a standard deviation of 1.69 for Trial I, and 1.36 with a standard deviation of 1.34 for Trial II. After this point, some decrease was observed but frequencies were low.

Correlations between trials ranged from .713 at the 50-99 millisecond class interval to .123 at the 250-299 class interval. The median correlation was .270 which occurred at the 550-599 millisecond class interval. To test for a significant difference between means at each class interval, t tests for related measures were used. No significant mean differences were obtained. The moderate to low correlations indicated some tendency for these subjects individually to replicate.the temporal properties of their speech events from trial to trial. The lack of significance observed by t test procedures indicated that these subjects as a group replicated durational. properties of speech events.

Frequency Distributions of Silence Events. In this research, silence events were defined as periods occurring during connected speech where the noise level of the apparatus









28

was not transcended and possessed a duration in excess of 50 milliseconds. The means, standard deviations and correlations between Trial I and Trial II for silence events of fluent speakers are presented in Table 2. The mean frequency of silence events between 50-90 milliseconds for Trial I was 103.02 with a standard deviation of 22.08. For Trial II the mean frequency of silence events was 106.36 with a standard deviation of 17.90. The mean frequency of silence events between 100-149 milliseconds for Trial I was 60.02 with a standard deviation of 17.27 and for Trial II, a mean of 58.47 with a standard deviation of 13.81. The mean frequency of silence events at the 150-199 millisecond class interval for Trial I was 24.48 with a standard deviation of 10.21 and for Trial II a mean of 22-78 with a standard deviation of 9.11. This decrease in mean frequency of silence events continued through the 400-449 milliseconds class interval where for Trial-I the mean frequency was 2.81 with a standard deviation of 1.92. The data remained asymptotic until the fifteenth interval, 750 milliseconds and beyond, where for Trial I the mean frequency was 9.55 with a standard deviation of 4.75 and for Trial 11 8.67 with a standard deviation of 4.35. Because this class interval had no upper limit, it was not comparable to the fourteen proceeding class intervals.








29


Table 2. Means, standard deviations and correlations of
Trials I and II for frequency of silence events of 64
fluent speakers.
Time in Mean Standard Correlation
milli- Deviation
seconds Trial I Trial II Trial I Trial II r I II

50- 99 103.02 106.36 22.08 17.90 .624
100-149 60.02 58.47 17.27 15.81 .601
150-199 24.48 22.78 10.21 9.11 .474
200-249 9.23 8.56 6.02 4.96 .297
250-299 4.84 4.17 3.39 3.84 .501
300-349 3.63 2.91 2.64 2.70 .209
350-399 3.16 2.77 2.50 2.13 .587
400-449 2.81 2.59 2.01 1.92 .338
450-499 3.33 2.88 2.02 2.23 .199
500-549 2.55 2.58 1.78 1.79 .228
550-599 2.53 2.05 1.81 1.65 .221
600-649 1.91 2.17 1.83 1.60 -.042
650-699 2.13 1.73 1.70 1.44 .137
700-749 1.61 1.63 1.22 1.50 .197
750-X 9-55* 8.67* 4.75 4.35 .806
xI and xii differ beyond .05 level of confidence.









30

Correlations between trials ranged from -.042 at

the 600-649 milliseconds class interval to .806 at the 750 milliseconds and beyond class interval. The median correlation was .297 at the 200-249 milliseconds class interval. To test for a significant difference between means at each class interval, t tests for related measures were used. For the frequency of silence events above 750 milliseconds, the means were significantly different beyond the .05 level of confidence.

Comparison of Fluent and Disfluent Speech Samples

Contingent upon the observed stability of the mean

estimates for frequency distributions of speech and silence events for fluent speakers, it was hypothesized: The durational data for disfluent speakers will deviate from the mean estimates obtained for the fluent speakers at one or more class intervals.

Oral Reading Time and Disfluency Frequency. The mean reading time for the disfluent speakers (stutterers) was 229.25 seconds with a standard deviation of 168.78 as compared to a mean reading time of 92.62 seconds with a standard deviation of 11.80 for the fluent speakers. To determine the significance of the difference between these groups, a t test for independent measures was calculated. The results indicated the mean difference between these groups was significant at the .05 level of confidence.









31

The mean disfluency frequency for the stutterers was 42.33 with a standard deviationof 44.12 as compared to 3.94 with a standard deviation of 2.61 for the fluent speakers. The results of a t test for independent measures indicated the mean difference between these groups was significant at the .01 level of confidence.

Frequency Distributions of Speech Events. The means, standard deviations and t tests comparing the :frequency distributions of speech events for stutterers and fluent

speakers are presented in Table 3. Of the f ourteen, 50 millisecond class intervals for speech events, two were observed to be significant beyond the .01 level of confidence. These intervals commenced with 150 and 200 milliseconds. Three additional class intervals were observed to be significant beyond the .05 level of confidence. These were the class intervals commencing with 100, 250, and 350 milliseconds. The remaining class intervals of speech events failed to discriminate between the fluent'speakers and the stutterers. Where significance was observed, the mean frequency of speech events was higher for the stutterers than for the fluent speakers.

FrequencZ Distributions of Silence Events. Table 4 presents the means, standard deviations, and t tests comparing the frequency of silence events for stutterers as a group with fluent speakers. As observed from this table,








32

Table 3. Means, standard deviations and t tests comparing
12 stutterers with 16 fluent speakers for frequency of
speech events.

Time in Mean Standard t of x 1milli- Deviation x2
seconds
Stutterers Fluent Stutterers Fluent Speakers Speakers
50- 99 76.42 62.75 38.32 13.87 0.62
100-149 70.25 46.31 29.07 9.06 2.65*
150-199 60.67 41.69 15.19 5.64 395**
200-249 50.17 30.06 19.03 6.87 3.35**
250-299 32.58 19.63 16.09 5.34 2.57*
300-349 25.08 16.63 12.84 6.18 2.02
350-399 19.42 9.69 12.61 3.03 2.51*
400-449 13.75 8.38 11.40 4.70 1.47
450-499 7.83 4.56 6.63 3.27 1.51
500-549 7.17 2.94 6.78 1.69 2.02
550-599 4.92 2.06 5.29 1.95 1.71
600-649 3.17 1.56 3.89 1.79 1.28
650-699 2.17 1.25 3.67 1.57 0.52
700-749 1.58 0.94 1.38 .93 1.35
750-X 2.58 1.31 4.07 1.74 0.64
t .05 (df = 26) = 2.06

t .01 (df = 26) = 2.78








33

Table 4. Means, standard deviations and t tests comparing
12 stutterers with 16 fluent speakers for frequency of
silence events.

Time in Mean Standard t of xlmilli- Deviation 2
seconds
Stutterers Fluent Stutterers Fluent Speakers Speakers
50- 99 107.42 101.75 44.20 27.58 0.37
100-149 63.83 55.56 18.68 17.82 1.14
150-199 39.50 22.51 18.37 9.24 2.85**
200-249 22.67 8.94 15.58 3.89 2.85**
250-299 12.92 5.19 8.40 4.46 2.78**
300-349 10.92 3.25 8.28 2.27 2.98**
350-399 7.25 3.51 6.48 2.47 1.91
400-449 7.67 2.38 5.80 1.96 2.91**
450-499 6.25 3.06 5.48 1.98 1.84
500-549 6.75 2.69 5.45 1.82 2.37*
550-599 5.42 2.44 4.25 1.93 2.18*
600-649 4.85 1.38 4.16 1.09 2.70*
650-699 6.08 2.13 5.50 1.78 1.70
700-749 4.67 2.00 4.19 1.32 2.05
750-X 52.25 9.13 16.68 4.11 2.58*
t .05 (df = 26) = 2.06
t .05 (df = 26) = 2.7806
**
t .01 (df = 26) = 2.78









34

four of the fifteen class intervals were significant beyond the .01 level of confidence. These intervals commenced with 150, 200, 250, and 300 milliseconds. An additional four class intervals were significant beyond the .05 level of confidence. These were the intervals commencing with 500, 550, 600, and 750 milliseconds. In each significant class interval, the mean frequency of silence events was greater for the stutterers than for the fluent speakers. Effects of Time of Da and Ex-erimenter-ubjct Combinations

Two purposes of this research were to test the

following hypotheses: (1) The time of day during which subjects are run does not significantly influence frequency distributions of speech and silence events, and (2) The sex of the experimenter in combination with sex of the subject does not significantly influence frequency distributions of speech and silence events.

A two-dimensional analysis of variance was used,

treating time of day and sex of experimenter in combination with sex of subject as main effects. An analysis of variance was performed for reading time, disfluency frequency and each of the fifteen class intervals f or speech and silence events.

For the four analyses of variance treating the effects of the experimental variables on oral reading time and dis-









35

fluency frequency, non-significant F ratios were uniformly observed.

Sixty separate analyses of variance were obtained for the class intervals of speech and silence events. Six significant F ratios at the .05 level of confidence were obtained for speech events and six significant F ratios at the .05 level of confidence were obtained for silence events. The data were inspected for potential patterns of significance. Each significant F ratio failed to replicate itself with respect to significance level on its comparable trial. Other patterns considered were significance by trial and significance by variable. Such inspections failed to discern a rational pattern of significance. From the obtained 180 F ratios, by chance, nine may be expected to be significant beyond the .05 level of confidence and two significant beyond the .01 level of confidence. Therefore, the observed significances may reasonably be attributed to chance. Summary tables for these analyses are contained in Appendix B.














CHAPTER IV

DISCUSSION


The research findings of major importance were the significant differences observed between stutterers and fluent speakers for frequency distributions of both speech and silence events. It was hypothesized that because stutterers use more time to read a set passage than do fluent speakers, at least one class interval of speech or silence events would significantly discriminate between stutterers and fluent speakers. Five class intervals of speech events were found to significantly discriminate between groups of fluent speakers and stutterers. Nine class intervals of silence events discriminated between these groups#

The duratioual properties of vowels and syllables are reported to range from approximately 100 to 400 milliseconds (Black, 1949; House, 1961; Sharf, 1962). In the present research the 50 millisecond class intervals of speech events whose lower limits were 100, 1509 200, 250, and 350 milliseconds statistically discriminated between fluent speakers and stutterers. The durational properties of vowels and syllables may relate to observer defined sub-classifications 36









37

of disfluency which include interjections, part word repetitions, and word repetitions (Johnson, 1961; Young, 1961). At each significant class interval, the mean ftor the stuttering group was greater than for the fluent speaking group. The frequency distributions of speech events for each subject were determined by an automated durational analysis device. These findings suggested the vocal aspects of socially defined disfluency may be amenable to objective quantification which results from machine procedures. Socially defined disfluency may also possess acoustic properties which differ from fluent speakers (Bryngelson, 1932)i however, the above quantification applies only to the durational aspects of vocal emission.

The findings relative to speech events suggested that stutterers emit extraneous vocal behavior during oral reading. If the extraneous vocal behavior is temporally separated from other vocal behavior, it would be expected to be reflected in the frequency distribution of silence events also. Such temporal separations were indicated by the presence of nine significant class intervals of silence events which discriminated between the stutterers and fluent speakers. These were the class intervals whose lower limits were 150, 200, 250, 3005 400s 500, 550, 600, and 750 milliseconds. At each significant class interval, the mean of the stuttering group exceeded that of the fluent









38

speakers. Analysis of the preliminary research data (Appendix A) comparing stutterers to fluent speakers suggested the possibility of these findings. Roe and Derbyshire (1958) presented anecdotal material which supported their suggestion of the possibility of differences in this dimension between fluent speakers and stutterers. In addition to advancing the proposition that temporal separation between extraneous vocal behavior accounts for the elevation in the frequency count of silence events, there are other factors that may relate to the increase of silence events, Clinical note has been made (Williams, 195?) of an inhibitory or 'heel dragging' quality in the vocal behavior of stutterers. Wendahl and Cole (1961) reported listeners accurately discriminated between segments of stutterers' fluent speech and segments of fluent speakers' speech. They reported that the listeners relied in part on rate of speech to discriminate between the stuttering and fluent groups. if these findings are accurate, analysis of the frequency distribution for speech events of such edited samples of stuttered speech may not exhibit differences from segments of fluent speakers' speech. However, these stuttered speech samples with the moments of stuttering removed could be expected to exhibit differences in the frequency distribution of silence events, This proposition relies on the assumption that the durational properties of vocal behavior are fixed by the content of the communication.









39

Excessive prolongations of speech elements enter into the social definition of disfluency. Two propositions have been advanced to account for the elevation of the frequency of silence events for the stuttering group. These were temporal separation between extraneous vocal behavior emitted and differences in the temporal spacing of the acoustic element. The resolution of these propositions must rely on subsequent research.

A purpose of this research was to establish estimates of the reliability of frequency distributions for speech and silence events of fluent speakers. On a test-retest basis, sufficiently stable reliability was observed to permit group comparisons. While there was a tendency, with the procedures employed, for the fluent speakers to approximate the durational characteristics of their vocal behavior on a test-retest basis, the reliability coefficients were not of sufficient magnitude to permit accurate comparison of individuals to the group.

Group reliability was assessed by using t tests on a tlest-retest basis for the fluent speakers, For the fifteen class intervals of speech events, t tests for related measures indicated no significant differences between the means for Trials I and II. It will be recalled that the tri~als were separated by at least 24 hours. For fourteen class intervals of silence events, t tests for related








40Q

measures indicated no significant differences between the means for Trials I and II. For the class interval which counted the frequency of silence events of 750 milliseconds and beyond, significance was observed at the .05 level of confidence. Two possible explanations are tenable. From 30 t tests, at least one could be expected to be significant by chance. Secondly, the direction of the mean difference may-relate to the significant difference obtained between reading times for Trials I and II. For both reading time and frequency of silence events, the mean of Trial II was lower. This possible difference may support Ylinifie's (1963) conclusion that changes in reading time were largely accounted for by relatively few silence events of long durations. The lack of significance between Trials I and II for the other respective class intervals of speech and silence events indicated sufficient reliability for group comparison.

Assessment of individual reliability of speech and

silence events was made by calculating reliability coefficients for each class interval of speech and silence events. These reliability coefficients ranged from low to moderate values, As measured, it may be inferred there was a low to moderate tendency for these classes of vocal behavior to repeat when experimental conditions remained constant. The reliability coefficients do not approximate the necessary values (.90) to warrant prediction of individual behavior on these measures.








41l

Statistical treatment of the frequency distributions of speech and silence events for the experimental variables of the tine of day and experimenter-subject combinations indicated the observed differences could reasonably be attributed to chance. These statistical analyses indicated the experimenter-subject combinations and time of day variables as measured are not factors of measurable importance to the frequency distributions of speech and silence events. Implications for ADplication and Further Research

Clinically, Speech Pathologists are concerned with the behavior of stutterers as individuals rather than as a group. It was therefore appropriate to consider the data for the stuttering subjects individually. This was conveniently presented by utilizing Z score techniques. The formula used for converting the raw frequency distribution data to Z score form was as follows: x -x (T1cNemar, 1955). Where x stands for the observed frequency of stutterer's frequency of events at a given class interval, R stands for the mean of the fluent speakers for Trial I, and d6 stands for the standard deviation of the fluent speakers. The Z score procedure permitted an expression of an individual stutterer's deviations from the fluent speakeiOs means at the respective class intervals for speech and silence events.

For the speech and silence intervals which discriminated between the fluent and stuttering groups, the Z








42

scores were computed for the twelve stutterers and plotted in profile form. Cursory analysis of these profiles indicated the following: (1) when the disfluency frequency of these subjects was within the range of fluent speakers (Johnson, 1961) the Z scores for the significant speech and silence intervals were within three standard deviations of the mean for fluent subjects; (2) when the disfluency frequency was above that of the range for fluent subjects the Z scores for the significant speech and silence intervals were beyond three standard deviations of the means for fluent subjects; (3) with the more severe stutterers, the Z score deviations appeared to be proportional to the observed disfluency frequency; (4) within the stuttering subjects there were a number of different observer defined patterns of disfluency; and (5) the Z score profiles also appeared to exhibit differences among disfluent subjects. These data are presented in Appendix C.

It would be hazardous to be definitive about these individual analyses. The best available estimate of the standard error of these Z scores is the standard deviation of the group (McNemar, 1955). This is due to certain low coefficients of reliability observed in the previous portions of this research. Also, the disfluent sample was limited in number of speakers.

Refinement of the procedures employed will be necessary before an application of the technique indicated above








i'3

can be utilized with precision. Subsequent research questions should be directed toward evolving procedures which will produce reliability coefficients of sufficient magnitude. Achievement of high test-retest reliability will permit definitive statements regarding the durational properties of an individual's vocal behavior. A parsimonious first step towards this goal would be to investigate reliability coefficients as a function of the number of responses required from each speaker.

Assuming adequate reliability is achieved, the basic technique of automated analysis of frequency distributions of speech and silence events may offer support for the stuttering specialist. Such procedures could lead to rapid and reliable evaluations of disfluency which would not be subject to observer bias. The currently employed observer defined procedures for defining extent and type of disfluent behavior are extremely time- consuming. Also, when the Speech Pathologist is serving as z3oth evaluator and

clinician, bias re'Lative to progress w erapy is possible.

Automated machine T. cedures f or defi.z extent and type of disfluencies could be a valuable ad~~n., to clinical judgment.

Future research could provide a means to convert

norms for fluent vocal behavior to rate measures and program these norms into an electronic device which would provide








4L4

moment to moment statements as to the relative fluency of a speaker. Such an electronic device could be a useful tool for behavioral modification. Response contingent consequences for socially defined disfluency have been reported to increase or decrease the rate of disfluency depending on whether the disfluent behavior was reinforced or punished (Flanagan, Goldiamond, and Azrin, 1958 and 1959; Goldiamond, 1965; Siegel and Martin, 1965; Leach, 1966). Such an electronic device could then conceivably program reinforcement for fluent behavior.













CHAPTER V

SUMMARY


The purposes of this research were to obtain variance estimates of frequency distributions of speech and silence events which occur during oral reading of fluent speakers, to assay the reliability of these frequency distributions on a test-retest basis and to determine the effects of experimenter-subject combinations and time of day variables on these measures. A major purpose was to compare frequency distributions of durational properties of oral reading for stutterers and fluent speakers.

Tape recorded speech samples were obtained from 32 male and 32 female young adult fluent speakers on two occasions separated by at least 24 hours. These two trials provided the data for test-retest reliability estimates. The time of day and sex of experimenter were programmed in a manner to permit assessment of these variables on the frequency distributions of speech and silence events. Tape recorded speech samples were also obtained from 12 young adult stutterers.

Connected speech was treated as a series of alternating speech and silence events with varying durations. Speech events were defined as vocalizations which transcended 45









46

the noise level o f the recording apparatus and exceeded a duration of 50 milliseconds. Silence events were defined as periods during connected speech where the noise level of the apparatus was not transcended and exceeded a duration of 50 milliseconds. The speech and silence events which occurred during these tape recorded speech samples were counted and tabulated by their durations. These data were assigned to class intervals which permitted construction of frequency distributions of speech and silence events. A total of fifteen class intervals were used, fourteen having a period of 50 milliseconds and the fifteenth counting all events exceeding durations of 750 milliseconds.

The analysis of the results suggested that the following statements are warranted.

1. Variance estimates of*durational distributions
of speech and silence intervals have been
empirically demonstrated to have sufficient
reliability for group comparison.

2. Experimenter-subject combinations and time of day
variables were not factors of measurable importance with respect to reading time, disfluency
frequency, and frequency distributions of speech
and silence events.

3. Several class intervals of speech and silence
events were found to discriminate significantly
between groups of stutterers and fluent speakers.

The class intervals of speech events which significantly discriminated between groups of stutterers and fluent speakers were those associated with the durations of vowels








4?

and syllables. The class intervals of silence events which significantly discriminated between groups of stutterers and fluent speakers may relate to the temporal separation of extraneous vocal behavior which may be socially defined as disfluency. These significant silence intervals may also relate to the clinically noted inhibitory or 'heel dragging' effect reported for stutterers. The discussion considered possible implications for further research and application of these findings.



























APPENDICES













APPENDIX A

PRELIMINARY RESEARCH: DURATIONAL PROPERTIES OF SILENCE
INTERVALS DURING ORAL READING


This investigation was designed to obtain information concerning the durational distributions of silence intervals which occur during disfluent and fluent oral reading and to relate these data to disfluency and reading rates. Durational distributions are considered formal aspects of verbal behavior which can be made explicit by machine procedures. Procedures

General Procedure. Tape recordings were made of adult normal speakers and of stutterers as a passage was consecutively read six times. The subjects were selected to provide a wide range of reading and disfluency rates.

The tape recordings of the speech samples were played to an analysis device capable of classifying pauses into 33 durations, with the lowest duration being .05 seconds. The device cumulatively recorded the number of silence intervals by durational categories. The reading time per passage was obtained with a stop watch. Frequency of disfluency was recorded by an observer who listened to the tape recordings. The obtained durational distributions were plotted graphically and compared to disfluency frequency and reading rate.
49









50

A detailed statement of procedures and apparatus follows.

Selection of Subjects. Five normal-speaking college students, ages. ranging from 21 to 34 years, were selected to provide a base or what might be considered usual performance for this population. These subjects reported no training in speaking or oral reading with the exception of an introductory speech course.

Six subjects who had been diagnosed as stutterers by college speech clinics were also run. Their ages ranged from 20 to 27 years.

The rates of disfluency of stuttering subjects have

been reported to change markedly with successive readings of the same material (Johnson, 1937). The decrement in disfluency rate which occurs with successive readings of the same material, provided a rationale for each subject to serve as his own control with all conditions constant with the exception of trial number. Normally fluent subjects have also been reported (Starbuck and Steer, 1953) to exhibit attenuation of disfluency rate with successive readings of the same material. The differences in attenuation rate, however, were reported to be considerably fewer than those for stutterers.








51

Recording Conditions. Speech was recorded in isolated rooms regularly used for this purpose. The recording equipment was in an adjoining room. Recordings were made in the evening when only an experimenter and the subject were present.

Instructions to Sujcs The subject was seated at

a table before the microphone and shown the text he was to read. He was instructed that he would read the passage six times. When signalled, he gave his name, the trial number, counted to ten out loud, and then after a pause began

reaiZ lie was also jastruapted to take breaks only between readings of the passage and also to avoid rattling the paper or shifting his position. Answers to any questions regarding the purpose of the research were deferred to the end of the session.

Reading Passage. The reading passage was selected from The National Geographic and contained 1024 words. Sentences containing infrequently used names or words were eliminated. The passage was typed double spaced and each subject read from a ditto copy.

Definition of Disfluency. The experimenter listened to the recorded tapes twice and recorded each instance of disfluency for both runs. A disfluency was defined as any break, excessive pause, prolongation or repetition of the flow of speech.








52

Timing of Reading Rate. The time lapse from the

first to the last word of each reading of the:passage was timed manually using a .10 second stop watch. Time was rounded to the closest second. Five reliability checks indicated comparable times within .50 seconds.

Apparatus. An electro-mechanical device utilizing

relay attack time was used to time silence durations. When the voice-operated relay turned to the normally closed position, a count was recorded on the first counter and a current was applied to the coil of'the first of 32 relays. When this relay made contact through its normally open points, a count was recorded in the second counter and current was applied to the coil of Number 2 relay, and the process repeated for as long as the silence interval lasted or until all 32 relays were operative. Since relay attack varied with voltage, one voltage, 20 VDC, was used throughout. Attack time variability for each relay did not exceed .001 second. At this voltage, the characteristic attack time was close to .030 second. A disadvantage of the apparatus was that not all 33 class intervals were of the same duration.

All AC voltage inputs were run from constant voltage regulators which supplied 115 VAC. Thetape recordings were played on an Ampex 601, a single track machine at 7-1/2 inches per second. From the tape recorder, a 500 ohm output









53

was connected to a Hewlett Packard 5 watt 110 db attenuator, Model 350A. and then to an Allison Laboratory band pass filter, Model 2A. The filter was set to pass between 150 and 3600 cycles per second. The output of the filter was fed to a Hunter voice-operated relay and was monitored by a millivolt meter (10 mv full scale). The average peak intensities of speech for the various tapes were balanced by the attenuator to read between 2 and 3 millivolts on the meter. It was found that the speech intensity varied from recording to recording. No systematic shifts in speech intensity were observed during the reading of a passage. The sensitivity control of the voice-operated relay was set at the maximum sensitivity level. Measurement of attack release time of the voice-operated relay indicated .010 second attack time and .050 second release time.

Durations Used. A.20 VDC ground was connected to the common pole on the voice-operated relay. Its normally closed contacts were connected to a pulse former and to the coil of the first relay of the timing circuit. The pulse former required a .006 second pulse to operate and in turn delivered a .014 second pulse to the first counter. The Sedeco counters which were used required a .012 second pulse to operate. Hence, Counter 1 recorded all silence intervals over .056 seconds in duration.








54

When the s eay (Cae DDT telephone type) made contact normally pen, it delivered 50 VDC voltage through osne set of con-a.cts to the second pulse former which in Lurn fired Counter 2. The time di 'armnce between the activation of Counter 1 and Counter 2 as .5 second. Thus, Counter 2 recorded all silence intervals .er .091 seconds in length. Through the other set of contacts on Relay 1, 20 VDC was delivered to the coil on Relay 2. The process described would then repeat through the remaining relays and counters. The final counter recorded durations of one second or longer.

The analysis device was calibrated through the use of a timing circuit which delivered pulses of known durations. The calibration procedure was as follows. The voice-operated relay was replaced by the calibration timing circuit. Pulses of one second duration were then sent to the relay coil which was adjusted go that a pulse of one second duration would ust cause the last counter, Number 33, to record.
z~ l .. .s ch e v d th tj C 4c*].ib a-ti-o@ tiu in g ci cu it a

adjusted to produce a pulse in the tahge of the pred eijg counter, Number 32, until its firing time was established and the process then repeated for each of the remaining counters. Each day during the running of data, the last counter was recalibrated and several of the other counters were checked.









55

Reliability of the analysis device was established

by playing the same tape to the analysis device four times. Table 5 shows the results of dmrational distributions of silence intervals for four runs of the same tape-recorded passage. The first row represents the frequency of silence intervals which exceeded .091 seconds, and the last row representing the number of pauses that exceeded 1.050 seconds. The greatest difference in frequencies between successive runs of the sane tape did not exceed 5 per cent, except in those relays where the count was small.

__________ of Daa Each taeD@rord~d passage was

played through the analysis device twice. Accordingly, two durational distributions of silence intervals were obtained for each sample. The differences between the two samples were within the order of difference reported for the reliability of the analysis device. The mean of the two

runs was used to obtain a single datum point for each class interval of silence durations. Result s

Table 6 shows the oral reading rates for the normal subject by speaker and trial. A comparison of these data to a normative study (Darley, 1940) of the oral reading rate for college students provided an estimate of the normalcy of the reading rates of the subjects depicted in Table 6.








56

Table 5. Number of' silence intervals recorded at each
duration, when saute tape was run four times.
Relay Silence Run Run Run Ruin Difference % Error
Length 1 2 3 4
(msec)
1. 56 941 949 947 929 202
2. 91 706 701 719 705 18 3
3. 121 575 580 591 575 16 4
4. 153 433 423 438 420 18 45. 18344 340 349 334 15 4
6. 217 314 312 315 307 8 3
7. 245 299 300 298 292 8 3
8. 272 284 285 282 279 6 2
9. 300 276 276 274 274 2 1
10. 332 268 265 267 268 3 1
11. 363 260 257 259 262 5 2
12. 395 250 245 246 248 4 2
13. 428 236 236 237 239 3 1
14. 461 227 225 227 228 5 1
15. 495 217 214 215 212 5 2
16. 524 208 206 208 204 4 2
17. 559 191 188 195 192 5 3
18. 590 184 178 179 179 6 3
19. 622 171 166 166 165 6 4
20. 656 156 149 152 149 7 5
21. 690 143 157 141 139 6 4
22. 723 158 152 133 130 8 6
23. 755 128 123 124 120 8 7
24. 786 120 113 112 115 8 7
25. 822 109 107 109 106 5 3
26. 852 108 103 104 104 5 5
27. 881 102 99 99 100 3 3
28. 905 96 92 95 92 4 4
29. 929 92 87 90 90 5 6
30. 955 88 87 85 87? 4
31. 988 83 80 80 82 3 4
32. 1020 80 74- 73 77 7 10
33. 1050 76 71 71 73 5 7









57




Table 6. Oral reading rate for normal subjects expressed
in words per minute.

Subj ects Trials
I I III IV V VI Mean

1 159.5 168.9 164.7 164.3 165.0 161.3 160.6

2 160.0 160.0 163.1 165.2 165.2 162.7 162.7

3 160.4 188.9 192.7 189.1 198.8 198.8 188.1
4 161.7 178.1 179.6 178.6 188.9 191.1 179.7

5 190.5 190.5 187.9 186.2 190.5 187.3 176.0

Mean 162.5 177.3 177.6 176.7 181.7 180.2 176.0









58

The first trial of these data was used for comparison with Darley's norms. This trial represents the reading of over one thousand words per subject, while Darley sampled 300 words per subject. For Trial 1, a mean of 162.3 words per minute with a standard deviation of 16.3 was obtained. Darley reported a mean of 167.3 words per minute with a standard deviation of 16.2 for 200 college students. Considering the differences in exact content of the reading passages and the respective lengths of 1,000 and 300 words, the similarity appears more remarkable than the mean difference.

Inspection of the means by trials suggested a gradual increase in oral reading rate, 162.3 words per minute on Trial I to 180.2 words per minute on Trial VI. To evaluate the significance of this observation, a rank order Chisquare technique (Wilcoxon, 1949) was selected. The result of this analysis indicated this distribution of trials by subjects could have occurred approximately one out of five times by chance. Gibbons, Winchester and Krebs (1958) found that sustained oral reading does not result in statistically significant temporal variation but rather a uniform rate of speaking occurs.

Inspection of oral reading rate measures by subject suggested that subjects differed from each other although there was overlap.









59

The tape recordings of the subjects were played to an experimenter who marked disfluencies on a copy of the reading passage. Table 7 presents the percentages of disfluencies for the normal subjects by trial. The mean disfluency percentage for Trial I was 1 per cent with a range of 1/2 to 1-1/2 per cent. Norms (Johanson, 1961) for disfluency rates of fluent male college students were reported. A disfluency rate of 1 per cent was reported as typical during conditions of oral reading of a factual passage. With respect to this measure, the current sample was considered comparable at least in respect to central tendency.

Inspection of means by trial suggested a decrement in the percentage of disfluency with repeated reading of the passage. Starbuck and Steer (1953) reported this adaptation phenomenon for twenty-two fluent college students. To assist in an evaluation of these apparently collaborative findings of the present study, rank order Chi-square analysis of trials by subject was employed. This test indicated that decrement in disfluency rate could occur by chance less than .007 times. These findings suggested that the subjects in the current study were comparable with the fluent subjects studied by Starbuck and Steer with respect to progressive decrement in disfluency with successive

reading of the same material.







60





Table 7. Percent of disfluent words for normal subjects. Subjects Trials
I II III IV V VI Mean

1 0.927 0.732 0.244 0.390 0.244 0.195 0.455
2 0.927 1.025 0.927 0.488 0.390 0.634 0.732
3 1.367 0.976 0.732 1.025 0.683 0.537 0.887
4 1.464 1.269 1.123 0.927 0.781 0.341 0.984
5 0.439 0.634 0.292 0.048 0.292 0.146 0.308
Mean 1.025 0.927 0.664 0.576 0.478 0.371 0.673









61

To assay the significance of the range of disfluency exhibited by the respective subjects, Chi.-square statistical analysis was again employed. The results, probability of less than .002, indicated that the subjects exhibited disfluency rates different from each other and the ordinal relationship remained approximately constant through the decrement in disfluencies occurring with adaptation. Concerning the typical properties of these subjects, it should be noted that the range of disfluencies falls within the first to seventh decile of the norms for comparable groups (Joh~nson, 1961).

Analysis of the relation between oral reading rate (Table 6) and disfluency rate (Table 7) indicates the relationship for these fluent speakers is not strong. On Trial I, the fastest reader S-5 (190 wpm) had the fewest disfluencies (.5 per cent). The slowest reader, S-1 (139 wpm) emitted 1 per cent disfluencies, while the subjects in close approximation of the mean words per minute exhibited the greatest percentage of disfluencies. For stutterers, Sander (1961) reported a strong relationship between disffluency and oral reading rate (Pearson's r of .86). From the data presented here, it appears that for fluent subjects there is a greater independence between disfluency and oral reading rates.

Table 8 'Lists the oral reading rate in words per

minute for -the stuttering subjects by trial. For Trial I.








62






Table 8. Oral reading rate for stuttering subjects
expressed in words per minute.

Subjects Trials
I II III IV V VI Mean

1 63.8 78.8 84.9 93.2 92.7 101.2 85.8

2 79.9 108.0 113.2 116.8 113.6 103.8 105.9

3 98.0* 126.2* 132.1* 138.4*

4 102.2 124.9 124.9 114.2 90.2 94.4 108.5

5 103.6 116.1 118.4 132.1 132.7 129.6 122.1

6 140.6 177.6 181.2 169.7 168.8 171.6 168.3

Mean 98.0 121.1 124.5 125.2 119.6 120.1 118.1

Not used in computing means.









63

the words per minute ranged from 64 to 141, These values were within the first to sixth deciles for oral reading rate of 50 college age stutterers on a single trial of a 500-word passage (Johnson, 1961). The fastest rate observed, 181 words per minute for S-6, Trial III, compares, tUo the ninth decile. The mean oral reading rate for subjects over the six trials was 118 words per minute. This rate was between the fourth and fifth deciles reported by Johnson (1961). It appeared safe to assume the oral reading rates of these stuttering subjects were within the normative values presented for a similarily designated population. Inspection of means by trials suggested that oral reading rate increased anad then decreased as a function of successive readings of the same passage. To test the significance of this observation, reading rates were rank ordered by trial for each subject and submitted to Chi-square analysis. The results of this operation indicated the above distribution could occur by chance in approximately one out of five samples.

Table 9 lists the percent of disfluent words emitted by the stutterers, by subject and trial. The mean percent of disfluencies for Trial I, 10.3, fell between the fifth and sixth deciles as reported by Johnson (1961) and the range, 1 to 24 per cent, compared to the first to eighth deciles also reported by Johnson (1961) for college








64


Table 9. Percent of disfluent words for stuttering subjects.

Subjects Trials
I II III IV V VI Mean
1 23.83 16.41 12.79 8.98 8.79 7.42 13.04
2 12.21 6.05 5.37 5.08 5.08 3.91 6.28
3 14.16* 4.59* 3.32* 2.15*
4 5.47 4.39 4.59 5.47 7.52 7.71 5.86
5 8.87 5.37 4.39 3.32 3.42 2.64 4.67
6 1.02 0.63 0.83 0.88 0.54 0.93 0.81
Mean 10.28 6.57 5.59 4.75 5.07 4.52 6.13
Not used in computing means.









65

stutterers. It appeared that the disfluency rates exhibited by these subjects were within the range of the reported college attending disfluent populations. Examination of the means listed for the six trials suggested a decrease in disfluency as a function of trials. Inspection of the body of Table 9 indicated similar trends for S's 1, 2, 3, and 5. Subject 4' disfluency rate appeared to increase with repeated trials and S-6 did not appear to exhibit either trend. Chi-square analysis indicated no statistical statements were warranted (probability .20). While this observation appeared to be contrary to the frequently reported adaptation phenomenon in the stuttering literature (Johnson, 1937), the position taken here is that the observation is consistent with the data. Further, differences in conclusions may be a result of the analysis procedure. Usual data processing involves reduction of measurements to mean and variance estimates. Using these procedures, the possibility exists that some members of the group will not follow the mean. Prior observation of these reversals in adaptation effect prompted the use of a statistical technique sensitive to heterogeneity within a group. Objective defense of this position was offered by Newman (1963) who reported six of twenty stutterers failed to exhibit decrement of disfluency with successive readings of the same passage,









66

Analysis comparing the relationship between oral

reading rate (Table 8) and disiluency frequency (Table 9) suggested results similar to those Sander (1961) reported. He reported a correlation of .86 between oral reading rate and disiluency frequency. Subject 1 had the slowest reading rate, 64I words per minute and the highest disfluency rate, 24 per cent, Subjects 2, 3, 4., and 5 were distributed between these extremes with two reversals. A statistical statement on this point was not feasible because of the limited sample size.

Figure "A presents the cumulative plot of the

durational distributions of silence intervals for normal subjects which occurred during the oral reading of a 1024 word factual passage. Data are collectiLvely presented for the five fluent subjects on the six trials. The data points for the trials which represent the upper and lower limits are connected by solid lines. Trials within these 'Limits are not connected. Cumulative plotting was necessary because zhe analysis apparatus did not permit equal class intervals. The upper limit of each class interval was represented, hence the first data point represents frequency of silence intervals between 56 and 90 milliseconds, the second between 91-120 milliseconds,- ------to 1050 milliseconds and beyond for the last data point. The curves for cumulative durational distributions for silence intervals of
















1300


1200 1100


S1000~
c7



to
























100



100 200 300 400 500 600 700 600 C) 0 1000



Figure 1. A composite plot of the cumulative frequency of silence intervals !or five fluent subjects reading a 1024 word passage six times Solid. lines connect the data points for the upper and lower limits. The individual data points falling between these limits are not connected.








68

subjects with the fastest oral reading rate were approximately the shape of an inverted L. As oral reading rate decreased, the inflection point was approached more gradually, with the curves of the slower reading subjects exhibiting a greater absolute frequency of silence intervals

at all durations,

Ninifie (1965) investigated the effects of instructions on durational distributions f or silence intervals, A group of twelve graduate students were instructed to read a 300-word passage in their (1) normal manner, (2) at a much faster rate, and (5) at a much slower rate. Timing of oral reading rate confirmed that the instructions produced significant differences in total reading time. Findings noted above relative to the point of inflection of the cumulative curve of durational distributions of silence intervals and systematic shifts in absolute frequency were observed. Mvinifie's research also pointed out the sensitivity of the durational distribution of speech and silence intervals to instructional variables. In the current investigation, where the instructions remained constant, cumulative curves of distribution of silence intervals for fluent subjects did not shift to an extent sufficient to warrant comment.

The data for durational distributions f or disfluent (stuttering) speakers were plotted for each subject









69

individually, comparing each subject to himself on several trials and to the composite durational distributions for normal speakers.

Figure 2 presents the cumulative durational distribution plot of stuttering S-i for Trials I, III, and V. The three trials have fewer silence intervals at the 90 and 150 millisecond class intervals than the normal group. At time intervals above these values the curve accelerates at an increasing and then decreasing rate relative to the normal group. Beyond 150 milliseconds, the curves for Trials III and V appear to approximate more closely those of the normal subjects.

Figure 3 presents the cumulative durational distributions of Trials I, III, and V for stuttering S-2. As in Figure 2, the frequency of silence intervals below 90 through 150 milliseconds is less than that observed for the normal subjects, although not to as great an extent as S-1. Above these time values, the curves accelerate at a decreasing rate compared to the curves -for the normal subjects. Trial 1. which exhibited a disfluency rate of 12 per cent and an oral reading rate of 80 words per minute, is the least similar to the curves for normal subjects, with the greatest differences occurring above 500 milliseconds. Trials IT7 and V which were similar in percentage of disfluencies and oral reading rateS, 5.4 and 5.1 Per cent










70










Percent of
13C0 Trial Disfluent Words S-I
S23.8
12C M0,I 12. 8

-'
I &8





103






C,














1 200 300 400 500 600 700 800 900 1000 TIME IN MILLISECONDS


Figure 2. The cumulative frequency of silence events for Subject 1 in Trials I, III, and V.






















Percent of
1.0 Trial Disfluent Words S-2
I 12.2 CI 54
5.1
1"1C L






70



















0 C O 20 300 4GO 500 600 700 800 9WX ICW TIME IN MILLISECONDS

Figure 3. The cumulative frequency of-silence events
for Subject 2 on Trials I, III, and V.








72

disfluencies and oral reading rates of 113.2 and 113.6 words per minute, exhibited similar curves of cumulative durational distributions for silence intervals.

The curves for the cumulative durational distribution of silence intervals for S-3 are presented in Figure 4p. Trial V is not presented because the noise level on the tapes for Trials V and VI was inordinately high. The preceding statements concerning the comparatively fewer silence intervals below 150 milliseconds are also observed here. The curve for Trial I shows an increase in silence intervals above 600 milliseconds, This increase occurred with S-2 on Trial I in the range of 500 milliseconds.

The curves in Figure 5 represent the cumulative

durational distributions for stuttering S-4 on Trials I, III and V. The curves for these trials exhibittographical properties similar to the previously presented stutterers. Relative to the normal subjects, they show an increasing and then decreasing rate of acceleration, with acceleration again increasing slightly in the 500 millisecond range. Also, for the class intervals ending with 90 through 150 milliseconds, there were fewer silence intervals. It is of further interest to note that this subject did not show a progressive decrement in disfCluencies with successive readings of the material. Reading rate increased and then decreased (Trial 1, 102 wpm, Trial 111, 125 wpm, Trial V,










73









Percent of 130 Trial Disfluent Words S-3
I--- 14.2
I- 3.3






I





















0
100 2 300 400 500 600 700 800 930 1000 TIME IN MILLISECONDS

Figure 4. The cumulative frequency of silence events
for Subject 3 on Trials I, III, and V.








74











Percent of 1300 Trial Disfluent Words S-4
I --- 5.5
1200 6

.4.6
Lbuco7 ~10C3
0










w-500

H 400




:52CO










for Subject 4 on Trials I, III, and V.









75

90 wpm). Disfluency percentage decreased and then increased (Trial I, 5.5 per cent, Trial III, 4.6 per cent, Trial V, 7.5 per cent). These cumulative durational distributions of silence intervals reflected these changes in as much as Trials I and V were more similar to each other than to Trial III, which had the fewest disfluencies and fastest oral reading rate.

The cumulative durational distributionsof silence

intervals for stuttering S-5 are presented in Figure 6. At the 90 through 150 milliseconds class intervals, the differences between the cumulative durational distribution curves for this subject, as compared to the composite curve for the normal subjects, are small. Trial I which was judged to have 9 per cent disfluencies, exhibited topographical properties similar to those of the previously presented stutterers. Trials III and V, where disfluency percentages were small (4 and 3 per cent), fell within the boundaries for the normal subjects.

Figure 7 presents the cumulative durational distributions of silence intervals for stuttering S-6 on Trials I, III, and V. This subject's curves were of particular interest because he was considered to be a stutterer both by himself and by several Speech Pathologists. Yet, during these trials his disfluency and oral reading rates were within the range for normal subjects. The differences below










76











Percent of U300 Trial Disfluent Words S-5







~1000
R. 9- 0


790





La
2c00






300




0
100 200 300 400 500 600 700 80 900 1000
-TIME IN MILLISECONDS__

Figure 6. The cumulative frequency of' silence events for Subject 5 on Trials I, III, and V.





















Percent of
1300 Trial Disfluent Words S-6
T ---l.0
o1200 TT .
.1- 0.8
0.5
n1100



00

9800



do
.,,J









730 OO0'













0
100 200 300 400 500 600 700 M0 M0 100 TIME_ IN MILLISECONDS


Figure 7. The cumulative frequency of silence events
for Subject 6 on Trials 1, 11, and V.








78

150 milliseconds were not remarkable when comparing this subject to the normal subjects. The general topographical properties for Trials I, III, and V of this subject more closely approximated the durational distributions of silence intervals for the normal subjects than for the stuttering subjects previously reported.

To varying degrees, the durational distributions of silence intervals for stutterers appeared to differ quantitatively from the composite data for normal speakers on the first several class intervals, 90 through 150 milliseconds. The stutterers had fewer silence intervals at these durations than did the normal speakers. There appeared to be a tendency for the curves associated with the highest disfluency'percentages and slowest oral reading rates to be most dissimilar from the composite curves for the normal subjects. These curves showed a tendency to increase and then decrease in acceleration as compared to the curves for the normal subjects, As the disfluency percentages and oral reading rates of the stutterers more closely approximated those of the normal subjects, the curves for the cumulative durational distribution of silence intervals more nearly approximated those of the normal subjects. The cumulative durational distribution curves for a stuttering subject whose oral reading and disfluency rates were within normal range were essentially within the range of the normal subject's cumulative durational distributions,









79

The observations herein reported seem to warrant

further research of speech and silence durational distributions as potential areas which mayr lead to objective procedures for measuring the type and extent of the disfluency phenomenon,













APPENDIX B

SUPPLEMENTARY STATISTICAL SUMMARIES


Table 10 presents the reliability coefficients for the Multiple-Class Time Analyzer, Table 11 presents the reliability coefficients for the reading time and disfluency measurement procedures.

Tables 12 through 15 present the summaries of the

analyses of variance for experimenter-subject combinations by time of day variables for reading times on Trials I and II and for disfluencies occurring during Trials I and II.

Tables 16 through 27 present the summaries of the significant analyses of variance for experimenter-subject combinations by time of day variables, Tables 16 through 22 present these siimma ies for speech and silence events occurring during Trial I. Tables 23 through 27 present these summaries for speech and silence events occurring

during Trial II.










80








81
Table 10. Reliability coefficients of the Multiple-Class
Time Analyzer for speech and silence events. Time (milliseconds)
CI Speech Silence

50- 99 .978 .995
100-149 .969 .992
150-199 .955 .986
200-249 .944 .957
250-299 .948 .976
300-349 .950 .932
350-399 .947 .970
400-449 .970 .938
450-499 .971 .844
500-549 .917 .930

550-599 .934 .939
600-649 .951 .917
650-699 .960 .961
700-749 .963 .943
750-X .992 .988








82








Table 11. Reliability of reading time and disfluency measurement procedures.

Measure Correlation
Between Within

Reading time in seconds .986 .999

Disfluency frequency .974 .791








83

Table 12. Summary of analysis of variance for experimentersubject combinations by time of day variables for reading
time of Trial I..

Source d.f. SS ms F

E S Comb. 3 416.11 15.37

Time of Day 3 200.27 66.77

Interaction 9 511.22 56.80

Within 't8 L457.62 92.87








Table 13, Summary of analysis of variance for experimentersubject combinations by time of day variables for reading
time of Trial II.

Source d.f. SS ms F

E S Comb. 3 23.94- 7.98

Time of Day 3 177.64 59.21

Interaction 9 381.95 42.44

Within 48 39427.90 71.41








84

Table 14. Summary of analysis of variance for experimentersubject combinations by time of day variables for disfluencies occurring during Trial I.

Source d.f. SS ms F

E S Comb. 3 1.19 0.40

Time of Day 3 9.62 3.21

Interaction 9 106.38 11.82 2.00

Within 48 284.25 5.92









Table 15. Summary of analysis of variance for experimentersubject combinations by time of day variables for disfluencies occurring during Trial II.

Source d.f. SS ms F

B S Comb. 3 6.05 2.02

Time of Day 3 18.17 6.06 1.21

Interaction 9 83.27 9.25 1.85

Within 48 239.75 4'.99








85

Table 16. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of silence events between 100-149 milliseconds
on Trial I.

Source d.f. SS ms F

E S Comb. 3 3,032.69 1,010.90 3.83*
Time of Day 3 338.94 112.98
Interaction 9 2,773.18 308.13
Within 48 12,652.17 263.59
F .05 (d.f. 3 5/40) = 2.84







Table 17. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of speech events between 150-199 milliseconds
on Trial I.

Source d.f. SS ms F

E S Comb. 3 365.90 121.96
Time of Day 3 122.90 40.96
Interaction 9 622.29 2.94*
Within 48 1,124.92 23.45
F .05 (d.f. = 8/40) = 2.18








86


Table 18. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of speech events between 400-499 milliseconds
on Trial I.

Source d.f. SS ms F

E S Comb. 3 17.44 5.81

Time of Day 3 156.07 52.02 4.0l*

Interaction 9 37.81 4.20

Within 48 609.17 12.6.9

F .05 (d.f. = 3/40) = 2.84






Table 19. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of silence events between 450-499 milliseconds
on Trial I.

Source d.f. SS ms F

E S Comb, 3 19.69 6.56

Time of Day 3 8.44 2.81

Interaction 9 72.81 8.09 2.47*

Within 48 157.17 3.27
F .05 (d.f. = 8/40) = 2.18








87

Table 20. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of silence events between 500-549 milliseconds
on Trial I.

Source d.f. SS ms F

E S Comb. 3 5.07 1.69

Time of Day 3 28.57 9.52 3.17*

Interaction 9 30.06 3.34

Within 48 136.17 2.84

F .05 (d.f. = 3/40) = 2.84






Table 21. Summary of analysis of variance for experimentersubject combinations by time of day variables for the
frequency of silence events between 600-649 milliseconds
on Trial I.
Source d.f. SS ms F

E S Comb. 3 29.21 9.74 3.07*

Time of Day 3 8.21 2.74,

Interaction 9 21.60 2.40

Within 48 152.42 3.18

F .05 (d.f. 3/40)= 2.84









88


Table 22. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of speech events between 650-699 milliseconds
on Trial I.

Source d.f. SS ms F

E S Comb. 3 3.35 1.11

Time of Day 5 15.33 4.44 2,87*

Interaction 9 12.85 1.43

Within 48 74.42 1.55

F .05 (d.f. = 3/40) 2.84






Table 23. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of speech events between 250-299 milliseconds
on Trial II.

Source d.f. SS ms F

E S Comb. 3 25.57 8.52

Time of Day 3 154.57 51.52 2.92*

Interaction 9 358.44 19.81

Within 48 847.17 17.65

F .05 (d.f. = 3/40) 2.84








89


Table 24. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of speech events between 350-399 milliseconds
on Trial II.

Source d.f. SS ms F

E S Comb. 3 174.94 58.31 339*

Time of Day 3 66.57 22.19

Interaction 9 128.06 14.23

Within 48 824.67 17.18

F .05 (d.f. 3 5/40) = 2.84






Table 25. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of silence events between 650-699 milliseconds
on Trial II.

Source d.f. SS ms F

E S Comb. 5 1.19 .40

Time of Day 3 17.69 5.90 3.23*

Interaction 9 23.93 2.66

Within 48 87.67 1.83
F .05 (d.f. = 3/40) = 2.84








90


Table 26. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of speech events between 700-749 milliseconds
on Trial II.

Source d.f SS ms F

E S Comb. 3 0.82 0.27
Time of Day 3 2.07 0.69
Interaction 9 15.93 1.77 2.52*
Within 48 33.67 0.70
F .05 (d.f. = 8/40) = 2.18





Table 27. Summary of analysis of variance for experimentersubject combinations by time of day variables for
frequency of silence events beyond 750 milliseconds
on Trial II.

Source d.f. SS ms F

E S Comb. 3 186.19 62.06 3.47*
Time of Day 3 92.69 30.90
Interaction 9 53.06 5.89
Within 48 858.17 17.88
F .05 (d.f. = 3/40) = 2.84













APPENDIX 0

Z SCORE DATA FOR THE INDIVIDUAL STUTTERING SPEAKERS


Figures 8 through 19 graphically present the

individual Z scores for the significant speech and silence class intervals of the stuttering speakers.
































91