Group Title: differentiation of low fidelity circuitry by behavioral test response
Title: The differentiation of low fidelity circuitry by behavioral test response
CITATION PDF VIEWER THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00098355/00001
 Material Information
Title: The differentiation of low fidelity circuitry by behavioral test response
Physical Description: xii, 126 leaves. : illus. ; 28 cm.
Language: English
Creator: Smaldino, Joseph James, 1944-
Publication Date: 1974
Copyright Date: 1974
 Subjects
Subject: Hearing aids -- Testing   ( lcsh )
Speech perception   ( lcsh )
Audiometry   ( lcsh )
Speech thesis Ph. D   ( lcsh )
Dissertations, Academic -- Speech -- UF   ( lcsh )
Genre: bibliography   ( marcgt )
non-fiction   ( marcgt )
 Notes
Thesis: Thesis -- University of Florida.
Bibliography: Bibliography: leaves 121-125.
General Note: Typescript.
General Note: Vita.
 Record Information
Bibliographic ID: UF00098355
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: alephbibnum - 000582347
oclc - 14101286
notis - ADB0721

Downloads

This item has the following downloads:

differentiationo00smal ( PDF )


Full Text








THE DIFFERENTIATION OF LOW FIDELITY
CIRCUITRY BY BEHAVIO~rAL TEST RESPONSE













JOSEPH JAMIES`'SMALDINOO










A DISSERTATION PRESENTED TO THIE GRADUATE COUNCIL OF
THE UNIVERSITY O F LORIDA
IN PARTIAL FULFILLMENT OF THE~ REQUIREMENTS FOR THE.
DEGREE OP DOCTOR OF PHILOSOPHY






UNIVERSITY OF PLORIDA

1976



































UNIVERSITY OF FLORIDA
II1I 1111U 111I 1111111111111IIlII lIIIII
3 1202 08552 8536











ACKNOWLEDGMdENTS

The author gratefully acknowledges the criticism and

help provided by the supervisory committee during the formu-

lation, conduct and writing of this dissertation. Special

thanks is extended to Dr. Hutchinson and Dr. Sellers who

provided ea~uipment, materials, advice and unyielding support

throughout the study. Dr. Cutler and Dr. LaPointe, of the

Veteranls Administration i-oopital, contributed space, instru-

mentation, materials, time,:e..i i 1 comments and enth~usiastic
encouragement. Dr. Moore supplied' advice and encouragement.

Dr. Harris, of the Submarine Medical Research Laboratory,

Psychnophysiological Section, Groton, Connecticut, furnished

direction, editorial comments, materials and uncompromising

support.

Appreciation is extended to faculty, students and friends

who enthusiastically volunteered their time to act as subjects.

G~raditude is extended to Dr. Hiutton and Dr. Canahl at

the Veterains Administration Hiospital in Atlanta, Georgia for

permitting use of instrumentation and materials required for
conduct of the study.

Statistical advice was provided 'oy Dr. Rice of the Veterans

Admin~sitration Hospital, Dr. Shuster and Daryl Downing of the

Bi~ostatistics Decartment.

Dr. Yost of the Commuunication Sciences Laboratory provided

technical advisement for some of the electroacoustic measure-

ments .







Special gratitude is given to the author's wife, Sharon,

whose secretarial and artistic skills, and patience made this

study possible.

Especial thanks to his son, Matthew, without whose

constant and loving involvement this investigation would

have been completed weeks ago.










TABLE OF CONTENTS


ACK(NOL'EDGMENTS ........... .*********************** ii

LIST OF TABLES.......................************** vi

LIST OF FIGURES...............**.........***.**.*** viii

ABSTRACT............. 0...***************** ix

CHAPTER

1 INTRODUCTION ANID REVIEW OF' TRE
LITERATURE................................1

Statement of the Problem.............. 1
Review of the Literature.............. 2
Purpose of the Study. ........*******.. 11

2 MATERIALS AND PROCEDURES ...............,... 12

Electroacoustic Measurements.......... 12
Behavioral Measures................... 23
Preparation of Test Recordings........ 32
Subjects............................ 32
Experimental Design...:................ 34
Behavioral Measure Administration..... 34
Derivation of the Behavioral Scores
Assigned to Each Hearing Aid.......... 36

3 RESULTS AND DISCUSSION. .............. 0..... 37

Results.....************************ 37
Discussion of Results................. 69
Implications to Further Research...... 77

4 SUMMAYRY AND CONCLUSIONS................,. 80

APPENDICES

A FREQUEN:CY RESPONSE CURVES .................. 83

B STIMULUS MATERIALS COMT-PRISINJG TH!E
BEHAVIORAL MEASUES.......................,. 93

C INSTRUCTIONS TO SUBJECTS POR THE
BEHiAVIORAL MIEASURES ................... ..... 101










TABLE OF CONTENTS CONTIINUED


ArPPENDICES

D SAM~PLE SUBJECT RESPONSE F~ORMS.............. 109

REFE'RENCE S.............,...................... 121

MOCGRAPHICAL SKET2CH.......................r........... 126












LIST OF TABL~ES


TABLE Pgg2
1 THiE REFERENCE AND COMPARISON REARING
AIDS POR THE PAIRED COMiPARISON QUALITY
JTUDGM~ENTS..O.........................* 31

2 THiE SIGNAlL-TO-NOISE RATIOS AT WHICH
EACH BEHAVIORAL MEASURE WAS MIXED WITH
SPEECH SPECTRUM~ NOISE.....................* 33

3 THE IEARINIG AIDS THROUGH WHICH THE
EEHAVIORAL MEASURlES WSERE RECORDED FORI
EACH GROUP AND TH~E ORDERINGS OF THE
BEHAVIORAL IEASURE TEST ITEMS (A OR C)
FOR EACH GROUP....................******.. 35

4 THE PEARSON CORRELATION MAITRIX BETWEEN
THE ELECTROACOUSTIC CHARACTERISTICS AND
BEHAiVIORAkL MEASUiRES, AND THE LEGEND TO
ABBREV'IATIONS USED IN THE MATRIX.......,... 38

5 SUMM~ARY OF' THE FOUR HIGHEST PEARSO0N
CORRKELATPION COEFFICIENTS BETWEEN EACH
BEHAVIORAL MEASURE: AND THE ELECTR'.O-
ACOUSTIC CHARAC TERISTICS ................... 48

6 FACTOR LOADINGCS ON EACH OF THE: FIVE FACTORS
ON TPHE ELECTROACOUSTPIC CHALRACTERISTICS..*** 55

7 MULTIPLE REGRESSION ANALYSIS OP EACH
BEHAVIORAL MEASURE WITH THE FIVE
HIGHEST FACTOR LOADING ELECTROACOUSTIC
CHARAC TERIS T IC s**************************** 58

8 PEARfSON CORRELATION COEFFICIENTS FOR
EACH BEHAVIORAL TEST ANJD RETEST (QJ
THROUGH IM~DT8RT ARE GROUPED DATA).......... 61

9 MULTIPLE REGRESSION ANALYSIS OF THE
BEHAVIORAL MEASURES RETEST SCORES WITH THE
FIVE HIGHEST FACTOR LOADING ELECTROACOUSTIC
CiARACTERISTICS........................** 65










LIST OF TAELES CONTINUED


TABLE DggR

10 RANKC ORDERINGS OF THIE FIVE
CHARACTERISTICS DERIVED FROMd THE
FACTOR ANALYSIS FOR EACH BEHAVIORAL
MEASURE4 TEST AND RETEST..................... 68

11 THE BETA WEIGHiTS OF THE FIVE
BEHIAVIORAL MEASURES DETERMINED) TO BE
SENSITIVE TO E~LECTROACOUSTIC
INTERACTION DIFFERENCES, STABLE
AND RELIAELE................................ 74










LIST OF FIGURES


FIGURE Paegg

1 BLOCK DIAbGRAM~ OF THEE EQUIPMENT USED
FOR OBTAINING F'REQUTEN\CY RESPONSE OP
THE HEARING AIDS UNDER INVESTIGATION********. 13
2 FREQUENCY RESPONSE CURVE OF AID
NUMBER ONE USING THE 2CC Coupler..********** 14

3 FREQUENCY RESPONSE CURVE OF
HEARING AID TEST CHIAMBIER. ................... 15
4 BLOCK DIAGRAMIV OF EQUIPMENT USED FOR
OBTAINING QUADRATIC AND CUBIC
INTERM~ODULATION DISTORTION. ................. 21

5 INsTRUM~ENTATION USED TO RECORD
SPEECH DISCRIMINATION TEST THROUGH
THE H-EARING AIDS UNDER INVrESTIGATION. ....... 27
6 THE INSTRUMENMTATION USED TO PRODUCE
A 300 MSEC REFERENCE TONE, 200 MSEC
PAUSE AND 300 M~SEC: COMPARISON TONE
POR THE INTERM~ODULAITION DISTORTION TEST..... 28

7 BLOCK DIAGRAMI OF INSTRUMlENTATION EMPLOYED
TO ELErCTRPONICALLY M~IX SPEECH SPECTRUM NOISE
WITH THE BEHAVIORAL MEASURES AT VARIOUS
SIGNAL-TO-NOISE RAITIOS...............,.....,. 30


viii











Abstract of Dissertation Presented to the
Graduate Council of the University of Florida
in Partial Fulfillment of the Requirements for the
Degree of Doctor of Philosophy


TH-E DIFFERENTIATION OF LOW FIDELITY CIRCUITRY
BY BEHAVIORAL TEST RESPONSE

By

Joseph James Smaldino

June, 1974L

Chairman Edward C. Hutchinson
Cochairmana Daniel E. Sellers
Major Departments Speech


Audiologists are often called upon to recommend hearing

aids for their clients. The evaluative procedures used to

make these recommendations involve the clients' performance

on behavioral measures designed to estimate and quantify

speech intelligibility. It is documented in the research

literature that electr-oacoustic characteristics of hearing

aids can affect speech intelligibility, however, there is

disagreement as to which characteristics have the most pro-
found affect. Reasons conjectured for the disagreement

include (1) behavioral measure unreliability, (2) inadequate

assessment of electroacoustic interactions and (3) use of

behavioral measures which ar~e affected by different electro-

acoustic characteristics and/or interactions.

In order to test the validity of these conjectures

forty-two discrete electroacoustic measurements were made on







each of ten hearing aids. Ten behavioral measures (C.I.D.

W5-22 W:ord Lists (CID), V.A. Discrimination Test (VAH),

Modified Rhyme Test (MRT), a synthetic sentence test (SST),

C.I.D. Everyday Sentencess judged intelligibility (ESJ) and

write down response (ESW), a quality judgment passage (QJ)

and the Intermodulation D~istortion Test at three S/Nus

(IMDTO, -4, -8)) were recorded through the same aids, de-

graded with speech spectrum noise and presented to ten

normal hearing listeners.

Pearson correlation coefficients between subject test

and retest performances showed good reliability (r >.75) for

all of the behavioral measures except SST, ESW an~d IMDT8.

Criticism of these reliable measures on the basis of in-

stability of results, was, therefore, concluded to be un-

founded in these data.

Pearson correlation, factor analysis and multiple

regression calculations were performed between the electro-

acoustic characteristics and subject responses on the

behavioral measures, in order to test the soundness of the

other two conjectures.

Examination of the Pearson correlation matrix showed

that many of the electroacoustic characteristics were highly

correlated with one another. In these data, therefore, no

ch~aracteri~stic was a truly independent variable, but represented

the sum total of the moment-to-moment interactions with all

of the other characteristics. The supposition that inter-

action effects require a systematic appraisal seems warranted.

Assessment of solitary characteristics, apar-t from interactions







with other characteristics, was determined to be unrealistic

and, perhaps, misleading.

An obliquely rotated factor analysis was performed on
the electroacoustic characteristics to permit the use of a

subject-to-independent variable ratio appropriate for multiple

regression procedures. Five factors were extracted from the
data. The characteristic under each factor which had the

highest loading was selected as the single most influential
characteristic to that factor. Factor groupings on the

basis of high loadings, were labelled harmonic distortion,

maximum power output, bandwidth, gain and regularity of

frequency response.

Multiple regression analysis produced regression

coefficients (Ri), fhe square of which indicated how much

of the variance in each of the behavioral measures was

accounted for by the five factors, and beta weights, which

signified the relative affect of the five factors upon each
measure. On the basis of this analysis, a modified cross

validation and reliability coefficients, VANI, MRT, ESJ,QaJ

and IMDTO were determined to be stable and reliable estimates

of speech intelligibility. The beta weights, however, rank

ordered the influence of the five factors differently for

each measure. This result was interpreted to imply that each

of the investigated measures was influenced by different inter-

actions between thle electroacoustic factors.

Comparison of research results, based upon different

behavioral measures, was decided to be inappropriate and,

perhaps, misleading because the sensitivity of different







measures to electroacoustic interactions do not appear to be

equivalent.

Finally, an approach to hearing aid selection procedures

was proposed which uses knowledge of differential sensitivity
of behavioral. measures to electroacoustic interactions, to

predict speech intelligibility through a hearing aid.














CHAPTER 1
INTRODUCTION AND REVIEW OF THE LITERATURE


Statement of the Problem

Audiologists are often called upon to recommend hearing

aids for their clients. The evaluative procedures used to

makre these recommendations involve the client's performance

on behavioral measures designed to estimate and quantify the

intelligibility of speech (Ross, 1972).

It is well documented in the literature that electro-

acoustic characteristics of hearing aids can affect speech

intelligibility (Harris et al., 1961r Jerger et al., 1966a,bs

Olsen, JTabeley and Pappas, 1966p Jerger, 1967, Olsen and

Carhart, 1967r 01sen and Wilbur, 1967; Bode et al., 19681

Jerger and Thelin, 19685 Bode and Kasten, 19711 Olsen, 1971a,br
Witter and Goldstein, 1971, Smaldino, 1972,1973). Results

from these investigations, however, are La disagreement as

to which characteristics most profoundly affect intelligibility.

Among the reasons conjectured for the disagreement are in-

adequate appraisal of electroacoustic interactions (Harris et al.
1961f; Smaldino, 1972,1973), use of behavioral measures which

are affected by different electroacoustic characteristics and/or

interactions (Jerger and Thelin, 1967; Witter and Goldstein,

1971; Smlaldino, 19)72,1973) and behavioral measure unreliability

(Shore, Bilger and Hirsh, 1960t Ross, 1972).









Review of the Literature

The electroacoustic characteristics of hearing aids

which have been implicated as having ant effect upon speech

intelligibility include harmonic distortion, intermodulation

distortion, bandwidth, irregularity of the frequency response,

transient distortion and gain.

Behavioral measures which have been employed or appear

to have utility in hearing aid evaluative procedures include

monosyllabic word lists, rhyming monosyllables, sentence

lists, quality judgments and judged intelligibility.

The literature is replete with disagreement as to which

electroacoustic characteristics affect intelligibility as

estimated by different behavioral measures.

Hiaronic Distortion/Sentence Lists

In exploring the effects of signal-to-noise ratio,

frequency range, flatness of the frequency response curve,
harmonic distortion and intermodulation distortion, Harris

et al. (1961) used the C.I.D. Everyday Sentences which were

equated for length (6-9 words per sentence) as the behavioral
measure. Harmonic distortion was found to have the highest

correlation with speech intelligibility error scores. Inter-

modulation distortion also correlated well -with intelligibility,

but was highly intercorrelated with harmonic distortion, thus,

was not considered a different variable from harmonic dis-

tortion. Cubic and quadratic intermodulation distortion were

found not to affect intelligibility. Frequency range was

slightly correlated with speech understanding, while integrated






area under the frequency response curve was found to be

moderate predictor of intelligibility. Transient distortion

was speculated as being the best portender of intelligibility,
but instrumentation for this measurement was unavailable to

the investigators.

Harmonic Distortion/Sentence Lists/Monos~yllabic Word Lists
Jerger, Speaks and Maalmquist (1966a,b) used a multiple
choice sentence intelligibility test (PAL Auditory Test,
number 8, Karlin et al., 194C) and three monosyllabic word

lists (C.I.D. Auditory Test, W-22, Hirsch et al., 1952r CNC

Word Lists, Lehiste and Peterson, 1959, PAL PB Word Lists,

Egan, 1498 which were low pass filtered at 500 Hz) to discover
a behavioral task which would reliably differentiate the

performance of hearing aids. Results showed that while the
sentence intelligibility task, with intellective masking,
rank ordered the aids in inverse proportion to harmonic dis-

tortion, "performance differences were not systematically

reflected in the monosyllabic word test results" (Jerger,

Speaks and Mvalmquist, 1966b, p. 253)-presented in quiet.
Harmonic Distortion/Rhyming M~onosyllables
Kruel et al. (1968) revised the M~odified Rhyme Test (MBRT).

The response paradigm consists of fifty closed set items with

six foils each. The subjects task was to draw a line through
the one fail that was heard. Bode et al. (1968) using the
MIRT found that consonant discrimination were reduced over

a ranpe- of thirteen to thirty percent with increasing harmonic

distortion (five percent, fifteen percent, twenty-five percent

and thirty-five percent).







Bode and Kasten (1971) using the Modified Rhyme Test

mixed with noise concluded that reduced high frequency band-
width, altered speech-to-noise ratio and harmonic distortion
in hearing aids were mutually inclusive causes of reduction
in consonant differentiations.

Inrtermodulati~on Distortion/MFonosyllabic Word Listis/Sentence Lists

Jerger (1967) found no consistent rank ordering of a
group of hearing aids when the C.I.D. W-22 Word Lists, PB
Word Lists, CNC Word Lists and the RaL-8 Sentence Intelligibility
Test were compared. The sentence intelligibility test did,
however, show reliable differences among the aids and rank
ordered the best aid first and the worst aid last. Another

test, The Intermodulation Distortion Test (IDT) based upon
distortion products that are generated when a 1 KHz and

1.6 K~e tone are mixed together was played through hearing
aids. The subject's task in a paired comparison paradigm

was to judge whether the signal passed through the.hearing
aid, or a reference signal made up of the same sinusoids not

paused through a hearing aid was heard first. Results showed
that the IDTP could easily differentiate hearing aids and that
it was superior to the sentence intelligibility test because
it had less inter-subject variance.

Bandwvidthf.~onosyllabic Word Lists
01sen and Carhart (1967) while investigating the use-
fulness of some test procedures for the evaluation of binaural
hearing aids, found that speech discrimination was reduced

when heard through a hearing aid, when compared with direct

reception at comparable signal-to-noise ratios. This' finding






led the investigators to explore the effects of bandwidth,

harmonic distortion and intermodulation distortion on speech

intelligibility. The behavioral measures were monosyllabic

words in quiet, monosyllables in the presence of competing

speech (Bell Telephone Intelligibility Sentences) and Mono-
syllables in speech spectrum noise. They found that bandwidth
was the only electroacoustic characteristic which consistently
rank ordered the aids the same way~ as the speech discrimination
scores

01sen and Wilbur (1967) assessed bandwidth, harmonic

distortion, intermodulation distortion and transient response
on a group of hearing aids. The behavioral measures recorded

through the aids were monosyllables in quiet and monosyllables
with intellective masking (Bell Telephone Intelligibility

Sentences). Conclusions drawn from the study were that band-
width was the only physical characteristic which ranked the

hearing aids in the same way as the speech discrimination

scores, and that monosyllabic words with competition was the

only measure which differentiated the hearing aids.
Response Irregularit~y/Sentence Lists

Jerger and Thelin (1968) assessed the effects of the

shape of the frequency response ourve, effective bandwidth,
harmonic distortion, gain, signal-to-noise ratio and signal-
to-hum ratio on speech understanding. The behavioral measure

employed wras the Synthetic Sentence Identification Test (SSI)
constructed by Speaks and Jerger (1965). The test consists
of a series of synthetic sentences based upon random or

conditional word probabilities. The sentences were approximations







to real English sentences, in that they had an appropriate

linguistic pattern, but little meaning. The response

paradigm of the SSI is closed set, wherein the subject merely
indicates which of a group of ten sentences was heard. A

high correlation was found between SSI and irregularity of

frequency response. The aids which produced the highest

speech scores, also had the smoothest frequency response curves.
A4n index of Response Irregularity (IRI) was devised which was

roughly proportional to the jaggedness or overall departure
from smooth uniform slope in the frequency response and which

showed the highest correlation with SSI scores than any other

investigated characteristic L They also found that the next

highest correlation with SSI scores was bandwidth below 1 KHz
and that harmonic distortion was "not a significant source

of degradation in speech understanding in the typical modern

hearing aid" (p. 172), indeed, SSI scores tended to be higher
in those aids showing the greatest distortion.

Int~ermodulation Distortion/M~onosyllabic Word Lists
01sen (1971a,b) found that aids having the least difference

frequency intermodulation distortion (CCIP Method) and broadest
bandwidth produced the highest discrimination scores (on a

monosyllabic word test with intellective competing message)

when persons with sensorineural hearing losses were tested.
In an attempt to determine whether the effect on discrimina-
tion was due to intermodulation distortion or bah~dwidth, a

study was conducted in which harmonic and intermodulation
distortion could be varied (using a peak clipper) and where

other performance characteristics could be held relatively







constant. Persons with sensorineural hearing loss and

excellent speech discrimination in quiet were tested with

monosyllables in quiet, with competing message and in the

presence of amplitude modulated white noise. Speech
discrimination scores in quiet remained unchanged even
with high levels of harmonic and intermodulation distortions

in the competing message condition, scores were slightly

improved a slight reduction in performance was noted in the
amplitude modulated white noise condition.

Transient Distortion/Quality Judgments

Witter and Goldstein (1971) correlated preference judg-
ments for short passages of continuous discourse recorded

through five hearing aids with harmonic distortion, inter-

modulation distortion, frequency range and transient distor-
tion values of the same aids. Transient distortion correlated

higher with subject preferences than frequency range or
harmonic distortion. Intermodulation distortion had little

effect on the quality judgments. In conclusion, the authors

stated that disagreement as to the important electroacoustic

characteristics to speech reported in other studies, may have
been caused by interaction of various characteristics. Transient

response, being an indication of overall linearity (interaction
effect) in an electroacoustic system, may therefore, be the

most appropriate single measure of the effect a hearing aid
will have on intelligibility.

Gain, Ban~dwidth, IRI, Transient Distortion/Sentence Lists
Smaldino (1972,1973) correlated thirty-one performance
characteristics of sixteen hearing aids with behavioral scores







on a multiple choice key word in sentences discrimination

tes-t (Kent State University Discrimination Test, Berger,

1969). Results showed that fifty-nine percent of the

variance in the discrimination test scores could be accounted

for by the combination of (in order of strength of effect)

white noise gain, bandwidth below 1 iEsi, bandwidth above

1 KHz, shape of the frequency response curve and transient

distortion. Conclusions included the hypothesis that inter-

active effects of performance characteristics may dictate

the moment to moment effect of the characteristics upon

speech understanding.

Quality Judgments, Judged Intelligibility and Revised
Monosyllabic Word Lists
Jeffers (1960) constructed eight, one minute selections

of continuous discourse and played them through five hearing

aids which represented a range of relative flatness of the

frequency response curve and gain. Subjects made paired

comparisons of the hearing aids, making a preference judgment
for the aid that sounded best to them. Results showed that

the electroacoustic characteristic differences represented

by these hearing aids did affect the quality of speech passed
through them and that subjects were excellent judges of the
differences.

Zerlin (1962) used a paired comparison technique with

six hearing aids. Subjects heard a thirty second long passage

of continuous discourse in a background of cafeteria noise.

Their task was to decide which passage of the pair was most

intelligible, or if equal, which wras the most comfortable to







listen to. C.I.D. W-22 half lists were recorded through

the same aids. Results showed that discrimination scores

derived from the W-22 half lists did not differentiate the

hearing aids, whereas, the preference judgments yielded

clear-cut and reliable differences for five of the six

hearing aids.

Speaks et al. (1972) suggested a scaling of intelligibility

based upon a decision by the listener as to how well continuous

discourse is understood. While the procedure has not been

used to differentiate hearing aids, the decisions reached by

the listeners correspond well with actual intelligibility scores.

Campbell et al. (1973) revised the C.I.D. W-22 W'ord

Lists. "About seventy percent of the original words were

replaced by words more suitable in familiarity and difficulty"

(p. 449). The revised lists have not been used to differentiate

hearing aids, although the presumed improvement of distribution

of difficulty over the original words suggest that they might

be useful in hearing aid selection procedures.

In summary, various electroacoustic characteristics of

hearing aids have been correlated with a variety of behavioral

measures designed to estimate and quantify speech intelligibility.

Research results are not in agreement as to which electro-

acoustic characteristics are most degrading to speech nor is

there agreement as to which behavioral measure to use to

estimate and quantify speech understanding.

For the purpose of this study measures of harmonic

distortion, intermodulation distortion, bandwidth, irregularity

of theI frequenyT response, maximum power output and gain were







employed. Behavioral measures included monosyllabic word
lists, rhyming monosyllables, sentence lists, quality judg-
ments and judged intelligibility.










Purpose of the Study
Results from investigations relating electroacoustic

characteristics of hearing aids to behavioral measures

designed to estimate and quantify speech intelligibility
are in disagreement. Among the reasons conjectured for

the disagreement are inadequate appraisal of electroacoustic

interactions, use of behavioral measures sensitive to different

electroacoustic characteristics and/or interactions, and

behavioral measure unreliability.

The purpose of this study was to explore the validity
of the reasons conjectured for the disagreement in research

results by investigating the effects of various electroacoustic

characteristics on several behavioral measures.

The specific questions asked wsere

What is the relationship between the electroacoustic

characteristics and the behavioral measures?

What is the relationship between the electroacoustic
characteristics?

What is the relationship between the behavioral measures?

Which electroacoustic characteristics are influential

in affecting the score derived from each behavioral measure

and what is the relative weight of the influence of each

characteristic?

What is the coefficient of reliability for each of the
behavioral measures?















CHAPTER 2
MATERIALS AND PROCEDURES



Ten used hearing aids of several different brands and

types were assembled for this investigation. Measurements

of frequency response, saturation output, gain, regularity

of the frequency response, bandwidth, harmonic distortion

and intermodulation distortion were made on each hearing

aid. The C.I.D. Auditory Test W-22, the Miami Yeterans

Administration Discrimination Test, the Modified Rhyme Test,

a synthetic sentence measure, a judged intelligibility test

using the C.I.D. Everyday Sentences, the C.I.D. Everyday
Sentence Test, the Intermodulation Distortion Test and

quality judgment material were all recorded through each of

the hearing aids and constituted the behavioral test battery.

Subjects' responses to these behavioral tests were correlated
with the electroacoustic measurements.



Electroacoustic Measurements


The electroacoustic characteristics of the microphone,

amplifier and receiver of a hearing aid interact so that the

gain across frequencies is not linear. Some frequencies are,

therefore, amplified to a greater extent than other frequencies.

A quantification of this frequency response is a graphic plot










FIGURE 1

BLOCK DIAGRAM OF THE EQUIPMENT USED
POR OBTABINING FREQUENCY RESPONSE OF
THE IrEALRING AIDS UNDER INVESTIGATION


B&K Type 4212 Hearing Aid Test Chamber
B&K' Type Condenser Microphone
2co Coupler
Hearing Aid Receiver
Hearing Aid Body
Speaker
B&K Type 1022 Beat Frequency Oscillator
B&K Type 2603 Microphone Amplifier
B&K Type 2107 Audio Frequency Spectrometer
B&K Type 3205 Graphic Level Recorder





14






FIG-URE 2





FREQUENCY RESPONSE CURVE OF AID NUMBER ONE
USING THE 2 CC COUPLER










RES ONSE ~l~--,~~tt~~~

dB -












80_
0~ 10 050 3 00

FREQUENCY IN Hz






































I


----t


---


RESPONSE
IN

dB


I~
Yii~---
II~I~F~II:


;_____


40I -





FREQUENCY IN Hz


-I


I


~ I


I---i--


I


t-;-H-i


(I


FIGURE 3





FREQUENCY RESPONSE CURVE OF
HEARINGS AID TEST C;HMBE~R


90_
-,_
-
=~li t -1-
R


-.-- ex r-


ii!


_ILILC; Ii '


----- H

;___I-. I--
-i
--







with the gain in decibels (dB) on th~e ordinate and frequency

in Herts (Hz) on the abscissa.

The procedure involves placing the hearing aid under test

in a Bruel and Kjaer (B&K) type 4212 hearing aid test chamber

(Figure 1). A 1 KHz tone was generated by a B&K type 1022

beat frequency oscillator and introduced into the test chamber

by a speaker. The receiver of the hearing aid was coupled

to a B&K pressure condenser microphone by a USASI standard

2 cubic centimeter coupler. The output of the coupler was

read in dB on a B&K type 2107 audio frequency spectrometer.

The voltage of the oscillator was set so that the free field

level of a 1 KHz tone in the test chamber was 70 dB SPL. A

second B&K condenser microphone, adjacent to the hearing aid

microphone acted as the monitor of a compression circuit made

up of a B&K type 2603 microphone amplifier and the oscillator.
The function of this cybernetic system was to keep the free

field sound level constant at 70 dB SPL at the face of the

hearing aid microphone. The gain control of the test aid

was set to a level 6 dB below its saturation output level,

res the 70 dB, 1 KHz reference tone. The oscillator was

automatically swept through the frequency range 10 Hz-10 KHE

and the relative gain of the aid for each frequency recorded

in dB versus Hz on a B&K type 3205 graphic level recorder.

This plot was called the frequency response ourve for each

test aid (Figure 2 and Appendix A).

Figure 3 shows the response of the B&tK 4212 test chamber
with a 70 dB input, and is included to provide a baseline







response of the system through which the hearing aid response

curves were obtained.

Saturation Output (Maximum Power Output)

"At anyr frequency a hearing aid has a limit to the

maximum sound pressure level that it can produce in the

test coupler, regardless of the input sound pressure at that

frequency. This maximum sound pressure level is called the
saturation output" (Berger, 1970, p.85).

The procedure for measurement includes placing the test
aid in the chamber and attaching it to a 2 co coupler as in

the measurement of frequency response. The voltage output

of the oscillator was increased until, with the aid's gain

control set full on, the output of the hearing aid did not

increase with a further increment in voltage. This point

was obtained for each aid and coupler at 500 Hz, 1 K(7z and

2 Kfs. The average of these three frequencies was termed

the saturation output of each aid.

Gain

Hearing aid amplification systems are not linear and

thus do not amplify different frequencies to the same extent.

The gain of a hearing aid is defined, under certain conditions,

as how much the hearing aid increases the output sound pressure

over the input sound pressure when the gain control is set

at maximum.

The procedure for measuring gain employed two different

input levels. The hearing aid under test was placed in the

B&cK equipment as described under frequency response. The gain

of the hearing aid was set at maximum. The input level for







a 500 Hz, 1 K~z and 2 K(Hz was introduced and monitored at

the face of the hearing aid microphone at 50 dB SPL and then

70 dB SPL. The difference between the input level and the

output level for each of the three frequencies was termed

the gain of the test aid. The average gain of each aid at
each level was the average of the gain at the three frequencies.

Index of Response Irregularity (IRI)

The IRI was measured as described by Jerger and Thelin

(1968) from the frequency response curve of each test instru-

ment. It is a quantification of the jaggedness or overall

departure from smooth uniform slope in an aid's frequency

response.

The procedure involved drawing a reference line parallel

to the frequency axis at the lowest reversal of the response

curve of more than 2.5 dB. Parallel lines were then drawn

at 3.0 dB intervals above this reference. The number of

crossings of these parallel lines with the response curve,

above the reference, was counted and termed the index of

response irregularity for that aid.
Bandwidth

Bandwidth is the range of frequencies over which a hearing

aid amplification system provides effective amplification. It

is specified by a low and high frequency limit of amplification

by several methods (Halpike, 1934; Lybarger, 1961 a,b,c;
Burnett and Priestley, 1964; USASI, 1960; Berger, 1970).

Hearing Aid Industry Conference (H.A.I.C.) Procedure.

H.A.I.C. bandwidth was obtained from the graph of the frequency

response curve (Berger, 19701 Lybarger, 1961 a,b,c). `The







ordinate values, in dB, for 500 Hz, 1 K1Jz and 2 K~iz were

averaged and plotted on the 1 KI~z ordinate, 15 dB below

the first. A straight line was then drawn through this

point parallel to the frequency a~ris. The low and high

frequency limits of amplification for that aid were the

frequencies where this line first intersected the frequency
response curve, moving in the direction of decreasing and
increasing frequency, respectively, from 1 K~z.

Houston Speech and Hearing Center (H.S.H.C.) Procedure.
H.S.H.C. bandwidth was obtained from the frequency response

of each hearing aid (Jerger and Thelin, 1968). A line was

drawn "parallel to the frequency axis at 10 dB below the

highest point on the response curve" (p.170). The low and

high frequency limit of amplification of the test aid was
defined as the frequency where the parallel line first

intersected the response curve, moving in the direction of

decreasing and increasing frequency, respectively, from 1 KHe.
Bandwidth calculations for both the H.A.I.C. and H.S.H.C.

procedures were expressed below 1 KHz, above 1 KHz and total.
Harmonic Distortion

If a pure tone of frequency f0 is passed through a li near
electroacoustic system, the output will contain only the fg'
with perhaps, phase and amplitude differences (Davis and
Silverman, 1970). If the electroacoustic system is nonlinear,

as are most hearing aids, harmonics of the input pure tone

frequency, ie., 2f 3f0'** 0g will also be present in the

output. Since the frequency range of a hearing aid defines







the high and low frequency limits of amplification and acts

as an effective filter, low fundamental frequencies will

generate more measurable harmonics than high frequencies,
because it is probable that more of the former are within

the effective handwidth of the hearing aid. The presence

of harmonic components of the f0 within the limits of
amplification of a hearing; aid is termed harmonic distortion.
The measurement procedure requires the same instrumenta-
tion described for procurement of the frequency response
curve (Figure 1). The gain of each hearing aid was set to

6 dB below its saturation output and the input signal was

always 70 dB SPL. The filter set on the B&K 2107 audio

frequency spectrometer was set to approximate one thire octave.
According to the International Electrotechnical Commission

Procedure (Berger, 1970) fundamental frequencies of 400 Hz,
1 KCIz and 1.5 K~z were used as input signals. The third

octave filters were centered on the first through third
harmonics of each of the fundamentals, ie., 400 Ezs 800 Hz,

1.2 KHz, 1.6 K~Xz; 1 KIzr 2 KHz, 3 KHz, 4 KF~z; 115 K~zr 3 K~z,

4.5 KHz, 6 KCHz and the energy present in dB recorded and
termed the harmonic distortion.

Difference Freouency Intermodulation Distortion

The instrumentation shown in the block diagram (Figure 4)
was used to generate and measure second (quadratic) and third

(cubic) order intermodulation distortion components. The
second order (quadratic) component was obtained using the
International Telephonic Consultation Committee (C.C.I.F.)
Method (C.C.I.F., 1937; Pete~rson, 1951). Two sinscidal test










FIGURE 4


BLOCK DIAGRAMI OF EQUIPMENT USED FOR OBTAINING
QUADRATIC AND) CUBIC INTERMODULATION COMrPON'ENTS


B&K Type 4212 Hearing Aid Test Chamber
B&K Type Condenser Microphone
2cc or Zwislocki Coupler
Hearing Aid Receiver
Hearing Aid Body
Speaker
B&K Type 2603 Microphone Amplifier
B&K Type 2107 Audio Frequency Spectrometer
Hewlet Packard Oscillators







signals (fi and fi +a 2) of equal amplitude (70 dB SPL)
were generated by two oscillators and mixed together. The
difference in frequency between the two oscillators was kept
constant at 400 Hz (Af). The mixed signal was applied to
the speakers in the hearing aid test chamber, where its

amplitude was monitored at 70 dB SPL by a B&;K condenser
microphone and type 2603 microphone amplifier. The output
of the hearing aid, located in the test chamber was applied
to a B&cK type 2107 audio frequency spectrometer and analyzed
by a one third octave band accept filter tuned to the difference

frequency Dlf (400 E~z). fl took the values of *S, 1 and 2 KE1z
for each aid investigated and correspondingly f2 took the
values of .9, 1.4 and 2.4 K~rz. The energy, in dB, observed
in the 400 Hz band was designated the quadratic difference

frequency intermodulation distortion.
The cubic intermodulation measurements were carried out

as above, except that the difference frequency measured at

the spectrometer was of the form 2f2 1 f. So that when fl
took on the values .5, 1 and 2 K~z and 2 took on the values
.9, 114 and 2.4 KHa, 2f2 was then 1.8, 2.8 and 4.8 K~iz and
the difference frequencies were 1.3, 1.8 and 2.8 KHE. The

dB output of the filters for these difference frequencies
was termed the cubic difference frequency intermodulation
distortion.









Behavioral Mleasures

Description of Mueasures

The following is a brief description of each of the

behavioral measures used in this study. Appendix B contains

the actual stimulus materials, Appendix C the instructions

given to the subjects prior to the presentation of each

measure and Appendix D is composed of sample subject response

forms.

Central Institute for the Deaf (C.I.D.) Auditory Test W-22.

The test is composed of several lists, each made up of fifty

monosyllabic words. The words were selected for familiarity

and phonetic balance, in that each list approximates the

phonetic composition of the English language (Hirsh et al.,

1952). Lists 1A and 10 were used in the study and represent

different orderings of the same words. Subjects were asked

to write down the word they heard. The intelligibility score

was denoted as percent correct.

Miami Veterans Administration Discrimination Test. The

items on this test were selected from the C.I.D. W-22 word

lists in order to improve equivalence among half lists and

to improve the distribution of difficulty of test items (Campbell

et al., 1973). Two half lists were chosen at random for

inclusion in this study. The lists were combined to form

one fifty item list. The fifty item list items were ordered

in two wrays employing a table of random numbers, which resulted

in list A and list C. Subjects were required to write down

the word they heard. Scores were expressed in terms of percent

correct.







Modified Rhyme Test (M.RT). The MRT is composed of six

different lists, each composed of fifty familiar American-

English monosyllabic words. The response sheets are of the
closed set multiple choice format with six foils in each

block. The word form in each block is consonant-vowel-

consonant, consonant-vowel or vowel-consonant. "La all cases

only a single initial or final consonant is varied, the
remainder of the word is consistent with its foils" (Kruel

et al., 1968, p. 538).

Modified Rhyme Hiearing Test number one was selected for

use in this study. The "a" foil was chosen as the stimulus

word. Two orderings from a table of random numbers were

labeled lists A and C. The subject's task was to cross out

the word in each block that they thought they heard. The

percent correct constituted the intelligibility score.

Synthetic Sentence Test. Three, ten sentence sets
(seven words each) constructed on the bases of conditional

probabilities of word sequences (Speaks and Jerger, 1965)

comprise the test. Only the third order set was used in
this study. Third order synthetic sentences were constructed

by drawing a word at random from a thousand word pool (Thorndike-

Lorge, 1944). A subject was asked to pick a word from the
same pool which could reasonably follow the first word in a

declarative sentence. The second word was given to another

subject to choose a third word using the same criteria e O*

In this way a second order synthetic sentence was constructed.

Third order sentences were based upon conditional probabilities

of word triplets. Word pairs were picked at random from the







second order sentences and a subject required to supply

a reasonable third word. The last two words of the created

triplet were given to another subject to supply a third

word, etc., until a seven word sentence was fabricated.

The ten, third order sentences were ordered in two

ways using a table of' random numbers and were designated

lists A and C. Subjects were required to write down the

sentence they thought they heard. A percentage score was

calculated from the number of correct words divided by the

number of possible words.

Central Institute for the Deaf (C.I.D.) E~verfVday

Sentencess Judgfed Intelligibility. Ten lists of ten sentences

each were constructed to represent everyday American speech

using specifications laid down by a Working Group (Grant
Fairboanks, Chairman) of the Armed Forces National Research

Council Committee on Hearing and Bio-Acoustics (Davis and

Silverman, 1970). Each list was formed with fifty scorable

key words.

The procedure employed in this study was suggested in

a paper by Speaks et al. (1972). Each subject was asked to

judge the intelligibility of each of ten sentences (0-100 percent)

and write the judged percentage on a response sheet. The

average of the judged percentages constituted the intelligibility
score.

Two orderings of list D were constructed using a table
of random numbers and labeled lists A and C.







Central Institute for the Deaf (C.I.D.) Everyday

Sentencess Write Down Response. Two orderings of list H

were prepared. using a table of random numbers and labeled
lists A and C.

Subjects were required to write down the sentences they

thought they heard. AI percentage score was calculated from
the number of correct key words.

Inte rmo dulat ion Dis~tortion Ts.Jerger (1967) reported

the development of a test based upon a subject's ability to

detect intermodulation distortion products.

Two pure tones (1 K~z and 1.6 KHz) were generated by
oscillators, monitored as to frequency using a frequency

counter and set to produce the same voltage output on a

vacuum tube voltmeter. The pure tones were combined (inter-
modulated) and tape recorded at VU=0. The tapes were played

through each hearing aid using the equipment shown in Figure 5

and designated the test tone. When these two sinusoids are

combined and transduced by a hearing aid, distortion products

are created at 400 Hz (2fl1 2Z), 600 Hzp(f2 1 f), 2200 Hz

(2f2 1 f) and 2600 Hz (1 + 2) (Jerger, 1967). These
products will be present at greater or lesser levels depending
on the hearing aid.
An identical intermodulated tone which was recorded on

tape but not played through the hearing aids served as a
reference tone. The reference tone and test tone were input

to the instrumentation shown in Figure 6. The output from

this equipment was recorded on tape and consisted of either

the reference or test tone of 300 millisecond duration, followed










FIGURE 5

INSTRUMdENTATION USED TO RECORD SPEECH
DISCRIMINATION TEST THROUGH THE
HEARING AIDS UNDER INVESTIGATION


B&K Type 4212 Hearing Aid Test Chamber
B&K Type Condenser Microphone
2cc Coupler
Hearing Aid Receiver
Hearing Aid Body
B&K Type 2603 Microphone Amplifier
B&K Type 2107 Audio Frequency Spectrometer
SONY TC 106A








F'IGURE 6


THE INSTRUMENTATION USED TO0 PRODUCE A
300 MSEC REFERENCE TONE, 200 MSEC PAUSE AND
300 MSEC COMPARISON TONE POR THE INTERMODULATION
DISTORTION TEST


Electronic Switch
One Shot
SONY TO 106A Tape Recorder
Audio Mixer
Matching Transformer
External Trigger
Oscilloscope







by a silent interval of 200 milliseconds and then either the
test tone or reference tone of 300 millisecond duration. The

order of tone presentation was arranged according to a table

of random numbers. Each comparison consisted of a reference

and test tone. One comparison was made for each of the ten

hearing aids under three signal-to-noise ratios.

The subjects' task for each comparison was to Judge

whether they heard the distorted or undistorted sound first.

A percentage correct score was calculated for each signal-
to-noise ratio.

Quality judgments. A thirty second long passage of
continuous discourse wcas played through each experimental

hearing aid (Figure 5). Paired comparisons were constructed

in which each hearing aid was the reference against which

all the other hearing aids were compared (Table 1). For

example, for group one, aid number one was the reference in
each binary paired comparison. Each aid was compared once

with each reference aid.

Subjects were asked to judge in each paired comparison

whether they thought the first or second paragraph was more

understandable. The total number of times an aid was judged

more understandable, divided by the total number of times

it appeared in a comparison resulted in a percentage score
for each hearing aid.











FIGURE~ 7





BLOCK DIAGRAM! OF INSTRUMENTATION EMPLOYED TO
ELECTRONICALLY MIX SPEECH SPECTRUM NOISE WITH THE
BEHAVIORAL MEASURES AT VARIOUS SIGNAL-TO-NOISE RATIOS


A SONY TC 106A or Wollensak 1520AY
Tape Recorder
B Grason-Stadler Model 162 Speech
Audiometer










TABLE 1


THE REFERENCE AND COMPARISON HEARING AIDS
POR THE PAIRED COMPARISON QUALITY JUDGMENTS




Group Reference Comparison
Aid Aids

1,11 1 2.3.4,5,6,7,8,9,10
2,12 2 1,3,4,5.6.7.89,10
3,13 3 1,2,4,5,6,7,8,9,10
4,144 1,2,3.5,6,7,8,9,10
5.15 5 1,2,3,4,6,7,8,9,10
6,16 6 1,2.,34.5,7.8,9,10
7.17 7 1,2,3.4,5,6,8,9,10
8,18 8 1,2,3,4,5,6,7,9,10
9,19 9 1,2.31,5.56.7,8,10
10,20 10 1,2.3.4,5,6,7,8,9









Pr~epartio of Test Recordings
The behavioral measures were recorded on magnetic tape

and played through each of the ten hearing aids using the

instrumentation shown in Figure 5. The tape recorded

measures were played by a SONY TO 106A tape recorder through

the speaker of a B&K type 4212 hearing aid test chamber. The

overall level of each measure was monitored at 70 dB RMIS at

the face of the hearing aid microphone in the chamber. The

output of the test hearing aid was applied to a condenser

microphone fitted with a standard USASI 2 cubic centimeter

coupler and fed into a E&K type 2107 audio frequency spectro-

meter. The spectrometer output was recorded by a second

SONJY TC 106A at VU=0.

All of the speech materials were recorded by the same

native English male speaker with no detectable regionalisms.

The recordings of the behavioral measures which were

played~through each hearing aid were input to a Grason-Stadler
Model 162 speech audiometer and electronically mixed with

speech spectrum noise at various signal-to-noise ratios

(Figure 7; Table 2). The output of the audiometer was recorded
at VU=0 and constituted the test tapes. A 1 KMz calibration

tone recorded at the same level as the behavioral measures

preceded each test battery tape.

Subjects
Twenty students and faculty at the University of Florida

served as subjects for this study. Each was screened for

normal hearing (in at least the right ear) res ANSI (1969).










TABLE 2


THE SIGNAL-TO-NGISE RATIOS AT WHICH EACH
BEHAVIORAL MEASURE WAS MIXED WITH
SPEECH SPECTRUM NOISE




S/N Ratio
CID W-22 Word List +8
vAH Miami Discrimination Word List +8
Modified Rhyme Test +2
Synthetic Sentence Discrimination Test +2
CID Everyday Sentences Judged Intelligibility +2
CID Everyday Sentences Write Down Response +2
Intermodulation Distortion Test 0,-4,
Quality Judgments +6







The age composition of the subjects fell within the range

20-35 years.

Experimental Design

Each subject listened to the hearing aid and behavioral

measure list shown in Table 3. All subjects listened to the

same Intermodulation Distortion Test. A reference aid was

assigned to each subject for the quality judgments as listed
in Table 1.

All subjects listened to the behavioral measure battery

twice. The second presentation took place- at least, five

days following the first and provided test-retest information.
The order of aid and list combinations were reversed from

those shown in Table 3 for the retest presentation.

Behavioral Measure Administration

Subjects were seated at desks in a room with low ambient
noise or in a small sound-treated test booth. If seated

outside the booth a noise attenuating muff was worn on the

nontest ear. Each subject was provided with a TDH-39 ear-

phone in a Mcx/rr1 cushion to the right ear, pencil and set

of response sheets (Appendix D).

The test tapes were played through a Wollensak Miodel

1520AY tape recorder. The volume of the recorder was adjusted

so that the 1 KCHz calibration tone which preceded each test

battery registered 80 dB SPL, through the earphone in a

standard 6 cubic centimeter coupler, on a Be&K sound level meter.

Prior to each test, the appropriate instructions were

read to the subject (Appendix C). The order of test presentation

was the same as the ordering of the instructions.











TABLE~ 3


THE HEARING AIDS THROUGH WHICH THE BEHAVIORAL ME~ASUREIS
WERE RECORDED FOR EACH GROUP AND THE ORDERINGS OF THE
BEHAVIORAL MEASURE TEST ITEMS (A OR C) FOR EACH GROUP


Group Aid and Behavioral
Measure Designation
1 1A,2C
2 2C,3A
3 3A,40e
4 4c,5A
5 5A,6C
6 60,7A
7 7A,80
8 80,9A
9 9A,10C
10 100,1A
11 1A,60
12 2C,7A
13 3A,80
14 40,9A
15 SA,1oc
16 80,1A
17 9A,20
18 100,3A
19 7C,5A
20 60,4A








Derivation of the Behavioral Scores
Assigned to Each Hearing Aid
Each hearing aid was heard by four subjects (Table 3).
The average score on each of the tests comprising the behavioral

battery was computed for the four subjects that listened to

each hearing aid. In this way a mean behavioral score on

each test was associated with every one of the hearing aids.

In the case of the quality judgments, the total number

of times each aid was judged more understandable was converted

to a percentage of the total number of times the aid appeared

in a comparison. This percentage was assigned as the quality

judgment score for every aid.

The Intermodulation Distortion Test was scored as the

percentage of times a subject made a correct choice through

each hearing aid. The number of correct choices was divided

by the total number of times the aid appeared in a comparison.













CHAPTER 3
RESULTS AND DISCUSSION



Results

The purpose of this investigation was to examine the

validity of reasons surmised for disagreement in research

results relating electroacoustic characteristics of hearing

aids and behavioral measures. The specific question,

statistical approaches and results were

W/hs~that ~ais the eaiosi Between the Electroacoustic
Characteristics and the Behavioral Measures?

Table 4 shows the Pearson product-moment coefficient

matrix for all electroacoustic characteristics and behavioral

measures. In Table 5 the four characteristics having the

highest correlation coefficients with each behavioral measure

have been arbitrarily chosen. These highlight the fact that

the highest correlations with each measure occur with different

characteristics and with different relative strength.

Measures of harmonic distortion, maximum power output, gain,

bandwidth and regularity of the frequency response achieved

the highest correlations with the behavioral measures. These

correlations were, in the majority of cases, negative.

What is the Relationshio Between Electroacoustic
Characteristics?

As shownr in Table 4 some characteristics are naturally

correlated because they are a measure of the same independent













TABLE 4




THE PEARSON CORRELATION MATRIX BETWEEN THIE
ELECTROACOUSTIC CHARACTERISTICS AND
BEHAIVIORAL MEASURES, AND THE LEGEND TO
ABBREVIATIONS USED IN THE M:ATRIX
























TABLE 4 CONTINUED



UPOA UPP1 <02 PAA CFas rel C,E2 GFA C9 SI

UoR L.1'0C08 0.07372 0.91441 0.935213 0.PFl?? 0. o2700 0.AO?37 0.=417 n,7ElOLa. c. Pac~
"n?1 C0."Y22 I,00000 0.*7951 0,arago O,69644 n.^5221 n.40175 0.55370 0.5170' C'lfor,
Mon C.10" 0.906>1 1.03 .rs120 n0,590 1 071o03 0.72450 0.71170 0. bP12L ?,*7 c;
Mo?A 0.03423 dheros9 0.1"l~a 1.00000 O76023 0.76767 0L7317 0.77061 0.el'ts 3.7L-<
';cq OotlZg Osn*49 0.R4cal 07al23 1.00000 C.75'96 P.72551 0.12010 1.ar30" C.C 23c
gas 0.'=770 0L122L 0.71sil 0. 7677 C190ma l.n3c00 0.o0912 0,rcasso C.e 12 spaCel
ge2 0. ??7 0 r0174 0."2440 CL'317 r.7'2RF 0.,0512 1.00?30 0.opnnn 1.'72(3 .0 712a
riwa 0.94<17 00'C'37 C.710 0.77-)1 raZ~lC 0.cqhns C02040 1.00?000 i.c321 C. c717Z
r.Se 07nlr^ 0. 1707 0.461'4 "*hLals 0.espon C~ans:12 077233 0,o3225, I1.0000 0.92~13
6S1 0,PenCY 0.01606 O.r'1, 9 0.7aceg n.aa2To o~oBISa 0.5"7l2 0."73ap Iep3 .Crrr 0
657 Pn30-a DefARER 9"5111 n,7o'5 07122' 0.li-33& 0.9"370 0ak727 019236~ 0.0 ~'9
Gca 005811l 0.^100' 0.FF Irr 0.761,2 0. 16C3a 0.'17605 t,80462 01?"310 ,CCpy;I )c-aga
*on r,'8710 0.C37"2 0.';. id .F6241 0,es~o0 0.1949 0.70023 0.95173 .n7 Tec Ccell 17
Ho C aa 0-1200 0.Aaste '),76#20 C,00000 C.502C 0.EAESb 0.=280< 1.8 sea 0.7i 01
,(712 0.779LI 0.~o~ OLAAn 0.A" 0'"02' O"TFER 0.4~9r07 0.5C'14 0.70127 0.reags 0.^u
-1')l5 C.'P=14 0.7^744 O.ants1 0,77?C0 r,96714 0.73o*2~ 0.62083 Os?1P IP ^7 71e I7L
Ir064 0.4170n 0 C.ea' C.Lsrla C.7F171 0.Calnln C9P323 0.7P1la C59751 0.P0371 E+C; 03
""1 C*4-'l" Osf#76 0 `7 0.'l? 0.7PinR 0.4?3 2 r.an?0R n~qo*54 0j.a77(0 gaLI'C CCr L
14r.3 0.la CE 0''0 0.74 DT 0. apan~ O.,*?? P10 I'ILTARD- F

Mfts no700a 0.rf 33 0. 31 ^*5;0 c.72< a 0.-413441 C,anI 25 ~( o0se P30
Iins C.Z0 7,an ~ o,75 .c~ Al' 0*27 .ll0 0.71ose $'7, 4 0.'cllo 0.".111, 0."O--
Inb~0 o.oot~ n, 2079 1. '973 n.,o-r^ n .'l3372 0.7sc63S 0.t^ .?C' 97 7llC
Lira ICn, *DM O ses) go 4502 0 .^1741 C.72o= 0,7075 0.n2"2 001003 0~li.7' O.li rg
0?<01 O. anal O.c7 .Lr2r 6 .r7l 42c2 ."TO 0.".'"21 C.Pslo0 0."na.20 I~ "Iom TP` i
01997 0.43aq pZpa OL"Inw l.c 7 (''7 001a 0.7405 n 37~ 0.47"7 G I-,
O~urea .r 21 0.PPIa 071 nrZrT, 1.~7 37 0I.clrrs= 0.'31 0.703093 0.0790.0r1
,rrnt EL~t? OF117 0.1 1 O.010s9 ..ro240 0.n000 0.40900 O,**Pr?9 0nuspa :OL o
01'""a 0.43'00'004 0.'S :'77 C^."?cOl r-,oc-L 0.~10395 0. 2700 0.1271 .o.oa 111J~
01Urt O.1317 .IlleaP7 -0.17206 -0.0127' r.1277; -0.0214 -0.11140 0.00CZC n,977pm .
0ATCA -0,17052 C0,tFlc -0.tid -0.^4^0 I'nl0 ~.I? -0.99207 -.00304 O,Croan 1,revo2 -
OH v'Cb 0.09138 0,03641 -0.1C17 -0.0 1 11 0.11 2n -0. 02083 -0.1961&S 0.001P '.131'I l

`IHOT -0.*0630 -0.2035 Oine12 -.3umes -0,472 -0.3nald0 -0.3712 -0,427?4 -.91024 -0.4I Ii
rvul C37 72 0.7nC04 0. I 017343 0.1 O 0,71 44 0.07097 0.19681 1.L7'11 071t ,;

rn~r -0-on 0 5-00.4101 I. 12 -.l'O -0 ^03 0.3 -0s"076' -0.7f73a -C,4. r
rnrM -0.'322T -O.47311 -0..ratta -0.lRO~II -0.749?~21 -oC0.1137 032 -0.5?1 27 -cOLP3 -"C.=a '3
warT -0.715es -07 4^R -0.30 '<3 -004370I -0,693.4 -C.45456 --0.10=70 -O.aR3AR -r.F .19 -.5 0
33TC -0.3RId -.7F -0 -0.011= -^n*2ap -;,mT1 -0.2006 0.110 -0.36s75 -O.4'770 --0.2? &P
r(s)C --0.44204~ -C3764 .0,gv47) wo.41.3337 00041 0020 -0.18170 0.2*7ay --g~irCe
Eln -0.2176?1 -C*DS -51 0..I'^^ -0,00~ 0.1420.0750 -02354 0.0ar-12 -0.10@,15 0.09 3
OJ-021m 0.7 -0-0.0241 -nr 1 -0.^iH20 -0244 ,os4 .3r08 *0.01~20 -r.rcn91 *0.07000
ucr' -Deagage -O9403 0.52 -0 J' --0.E1320 -r.1 5"I -0.?213 -.2 -0.19 -0.2025 0.1 3 -T.aE`-e

'0'4 -.^" -0*202 0.1I7 -0.7-10 .0.930 -02 ~~ 0aquY2 0.02001 -0,tal44 -Cl50^d -C.27ace

TurP9 -0.5004<, -0.F2395 -0.44021 -0.^s2po -0.247-,4 -G~c5745 -0.2s036 -0352A0 --0.1621a -0.a1+20




































G52 G 4 HDAb HOR rDI2 9 16


rrrP4 HD1 ~nz

0P170R 0.57155 ?.7F1P1
072snl n.r57r0 n.q~c~p
nC52~4 07P13~ r~3c0*
07hd7L 07nnR5 071LC3
Ca40PO 015~72 ~.~F-C~
0C232~ ODaZ~A g.~?ncE
07011" 0P"99n ORI'?C
0P7531 0;7717 0.9rZOn
01~1~1 0"1~70 0.'11797
0Yr903 00c~r7 o.?i~e~
C~e4*( OOFo2l L349~5
09~9*7 0.os17a 1.~2~~C
O~O0h' O'CLI~ n.7rnZS
004~0~ 01777"1 37nZ7~
Fn1157 0607h~ C73~1C
09nC~~ 07bG,4 07P'~~
L~~9~0 0"~J'3 091~~c
0.Pp143 LCCCC~ n.oIIrir
nP1~3a 001~55 IICC*r
0.753';3 00'r9' ?""I~d
0,77*197 I~~LS 0.13011
CIn2Ci7 01~912 O.lolCr
0.7~20a 0.9~~~~ O.~rg~~
074CCO 097~1~ OCL~~~
P71~b~ O.IC~n7 2Oa~Ua
Olaqoq 07'a?~ n.aalFI
C79111 096937 3"~EIC
0"1593 Oo'lnr. 1. oiF r-3r
DoP3d2 6.'3071) O~crdl
O'~ZJ~ 057~09 ~.T~17n
0r5771 0DL3~~ Ia"dOL
O.sl?~ OoFZO~ C^C~T~
(~"~:~91 Cra20Z 07'11)'~
0hO~h9 0~~IOF ~79o7P
O.R7pnR S"lhl' I*C 371
--CI~Tlo --O.Ca2~F --0.3c9sa
-013911 -0~aP37
-OLa~Th -009J31
-n.F~~Os -0n~E"~ -SC2~9C
-C20P~1 -C2.I~OU -09ZLC7
-0f~L37 -~4TFL? -~~PLOI
O~"aoO O~~CcS 01~2P'
-$n~l~10 -O.hs'00 -0.*l'"a
-07~2L3 -093871 '-Ct~rdOl
-C4491~ -05~901 -r.32210
-05~~~Z -0125~95 -~0719~
-0~63~0 -01~797 -02T~'P
-0L~107 Clll~l 0.11104
--C.agboh -00013~ -)0s^9~
--017032 -0.arPp5 --035326
-019~h~ --0IR~L5 -~IPn4C
-C21~97 --0~~P~2 -04Ch15


~b3

017'0e8
0t10~')
r),ri~c~
0711919
I777tP
OCIPC
O.,,~cg
C.p~~p
CI'IPP
Ce'70J
OcO~Oa
Cs,,Ln
0~"~22
P71"CC
0 7- '110
C71637
C7`3ET
~.~'?~7
C9'1~L
IO':ObO
r).c~ZLC
CCL~"O
C.c1',5
O rLZ~e
P"-~7'
IrC-?2C~
O.c42;1
0e14PI
C.P-C"7
C3~LrC
C7~CO~
Oclcc'
I71~C7
C.rilicC
OPclrs
-C.nc~C3
-"iPJIL7
-O~'~:Rr:
-Or I~~e
-0.~..1'0
-07POI')
0I"'OC

-011)11u
-C.3^5T~
O.GI~'OC
-OZ"Pe~
0C~34t
-01'102
-0L3F27
-I. l"loP
-0~1503


UD"5

rn~p
yp*l(
C.S5
G'I
T.~Z

G5rl
G9I
GC8
Gcd
*nb
u~n
~'D IP
Y'lh
H~II~



*33
unaa
~rrs
HC19
Y"3*
Hnb5
r[rr.
Hnnr
Olu?9
n:u~L
~ror
I!uns
rIUC5
C!Y"I
Clung
rI~r~
Udrb
Yh~rl
Hllrr
r9~(r~
HCl(il

~OI
C1?
Vb*
110~



CgX
9J
lunrLI
?*~'d
I~r~T1


p.n~~ca
n.sa4F~F.
O.Pe~nl
~~119EI
0lr227
0091~1
OnC1"
0"L~~7
~"23br
0~"7nr
IOPO~
O"~L~Z
(1413~*
0~pcc-r
P~~3'~
0~"1C1
C.~bzql
O~sZI
O~,ars
O."ncra
~.~-n-2
o.ri:s
r.i~r

0.sdo*l
P.rlc'3
Pc"" ~
C.,1?1'
0"42'1
CLILI~
ODlbr~
CC7CL"
C7271~
ClinsO
C.D4ai'
-0l'.l~CI
~0719~a
-C.ln4~-r
-O.asrbr
-il.3~EPF
-".a~SLq
IlI"TLa
~I~~L71
-01'1CO
-OIitZ72
~0LIZ9
-0111'1
01~'09
01~52L
-0b124C
-OIPCIZ
-01ngP04


0P5010
O.f~lCcS
~Lbr15
O'LlfZ
CclC.II
0''7906
C~cyq2
0OR310
r.rg2)Z
$ccLla
Oc~b-2
IP~eSO
Os~Ecl
OrlccO
C~IC31
~.`ILC4
Q.'cor~
o.rl!7a
n.o~l3~
~p17L4
~.FrP~3b
r.cbO\n
COI(:4

0'72~7
C'I~L5
OF7rbR
n.~71~0
0'b3'5
r).rz0ac

OCLPPI
L~Ln24
C'caS
nF~F-rO
-C~~19~
-n.(7p~q
-C.rE~~n
-p.sal~P

-OLIO~J
~.7ayra
-f .czso
-CSC3JO
-Ora~ll
-O.ICPqZ
-(I.ZD~1R
Oirb7?e
-0C571(
--0~20'7
-~L703o
-03BE3


0~~f71'3
OE1`C?
OEC2~r.
nf)511
O.'ircr,0
071~1i
0~'3~81
0.,9173
0F7'40

0'1~9h
OOzC~I
I.On~)3
051135

07~~89
09'1 r9
0. 13411
07'"PO
0170'72
O'i~'?C
"~~111
0.77nJa
0lb13h
0 ~,b.)0'
0.~nqir)
11'I)L4
0.:F~52
1~F'147
O.L,:ja
*11PO
r,.ncpl~
I~1Lrro
ChC330

0IL'15
C.ILsP

-0n~~5L
0.03i~J
-O ~SLCU
0 ieioo

-E~":94
-0.C4rTI
-OSarlP
-01911~
-002F.r~
-0L3r~e
-n.~1,72
-oLPlhs
-0177~1


070393
C7(irllO
O.~baeo
o.7ah20
.oo~o
a.r~iPo
C.~~qq~
"PZho5
~~E'rp*
0.77901
017PnSr
USIE'O
P'n1~5
L OC1OC
?.ocigs
n.~rsco
C.colOB
n7~7~L
07~173
~7P~FO
~7CC~O
~40~'6
07r'167
371\3Fb
07390C
07P 7r ~~
~7L9R7
CP33"1
O~R~OL
0l~lc
OC~~ZI
0'3007
0'lr1117
n4"n92
0PnL73
-"~Y~C.5
-01~267
-31^q77

-023~C~r
-0~3')1"
~.Cc)rn
--~'~~i49
-3' 1~22
-04.~405
--0~n07~1~
-069109.
-07n425
-~51503
-0114PO
-O~blOn
-OICZL7


0~779*1
O,ao*a
CC9111~
C7~~2~
(IP~--C
O1L'Z
PCE'I~
~.7a:21
(1.7a~?F
O~iO~lR
Li~~l"?
071(. 31
n.73'"F
0O~'C5
l.n00vC
0.Ooa-E
0cr,1"7
~C~7Cn
n.73"IC:
O.79'n1
n.-2i~
. rgn:IT
O.LOn7r
OLCII~
O.Lrr~l
07011LL
O.rS~~Z
07Lu7'~
07O'1.1
04~,17
C~31JT
0.7F71*
C"7"'1'
0~5~'1?
Cc 77 L7 I'
-~/ln.7
-".2'164
-C.Zr1~5
--n. c5 L)e
-0311C~~
-~hZ~l;
-".Eo'lrn
-Dcgl~,~
-0 7?o:(n
-0C12~'1'
-C~441Lil
-05~1LU
-O.Z~I*n
-C.F~-n2
-o. i.~~ 2
-C2~~',9
-02'C'B


C71~11
07h748
O.*4a5L
P.71a~0
PhT14
ir 710~2
O'ZOP1
C+319J
07~1714
07~~70
0~30~1
077416
0'971'1
0O1s~O
b~)"LSF
10901')
C"91'1'
0~1O~9
0~'")7
0711617
"~q750
r~0g10
Ci7~s5
07179L
Pc 7Zr~g
C~bf~2~
0171n"J
on01~5
r~~'94
Cljr.l 1
(.~"r,38
~.9113*
S.qn6+1
nT.pq~~
0*1017
-[121419
-073794
~tL907~
-0h7E71
~~. 1~4PS
-CF9121
r.~l~nl

-n.7134n
-09iq71
-e.~bhai
-051778
-C21a70
-c.saaor

-o.~25co
-o.?boai


TABLE .4 -CONTINUED


















TABLE LC CONT~NUED



HCIP HD19 Hn31: Hn~S MC HCIIC ntucc; orur I 4~vcp
072c72 O7aa7o O.r7~D~ 070CRI 0.~7pa0 C. I~r~s P~SDRI Oi~~~40 ~~eaSI 0'3~PR
ug~l CSSC~ O175~~ O.J~11 ".C")" nL77gC 0.4C~605 0.~~Pph ~'*P7C ~3'~n0
ynnp E~9~c~ 07011~ 09.c13 0*091 n.rblnr CrZO75 0*1102 O*"Sda Ur~*O* Oar~p~l
uonl C~I~Il F19~13 O.ql~Oj 0"3331 OII..SZI 0~la71 OFllrl 07u~SO d.12C~7 C~'IZ7
C'S e.'lbe~ 0~C~O~ 01'2on~ ~.r~g~a O.~~C~9 6.~.:~0 0.1.~4 n.n3~~a OC~rr;
6FI O~aal~ O+onj.~ OE~~rq O+'~i.9 74"'0 0'~372 071759 U?~I2~ "o~a21 0 ~q rla ri
b"7 n-1-1~ o.r,rr 0.~*525 o.~ilon n..n,L o.,soh~ 0"21?0 0.94490 I1.CfF"C 07~i)7
hrb 0~971-' 0011077 O~pyr.~ OsdnOP r75~10 0~~154 CPI.OJ ~Pe179 1~113~n CC~l'i~
C+5 C-"r~ n.Pa~~ 0"l~l~ O'IICr O.~Jsl? 0~3~r9 07*~7~ 30a~13 O.cnl~ CCL'P~
951 CPIL~~ OC~F~O OSD"C~ ll.971Ca n.p~~~~ n.7~537 ~CEC3b 09E37~ 503331 1).5e'"C
r.S? C~~07~ n.rclpe 0.: I In7 0 C(1S~5 ORdO~~l 0"1573 C9PRj7 001017 COEPi4 Er-L'I1C
'ljL Clh99n OI"LC45 0~1'~5 Os~~hl O.'ET.~ 0~lc15 OtA7onR 09'Lp0 C~d~.E CC2-24
~na O~arnO n.rs~dl 0. 77454 O.~C~O~ Oh.T~' O. h~CF~ 0."9roE r..rpdp C.'7~jp
Irna ~.7Lsa6 o.srpca 0.'n~rr 07.1.4 Pc 7141C C.7q7h~ 07nYr7 093J~I O~P.11 Ohl~l~
nl~lZ C~7rZ" 0.7'435 OIL?9,2 0r~~~7 OCIF11 o. '1441 0C7P7a 07~.Pn 1'r'SJ 044~17
ui* C'5~CO 07'23e ~717~1 (1.70~0 ~.,~C21 072130 o50L3C n7~~ua OCOCI I
P.7'dn7 CI'ZT~~ 0.73214 O.,C5t0 0.73341 D7~no9 0.7';113 0.8P59~ C.e~332
HSI C9LI?' CI 1012 0.11`00 0'74(P rO0Tn7 O.7CW73 O~~nJ' 0.~714h C.o3nrO OL7rCE
O.clC31 O~Plh~ 0.cr.c,, O"f~a~ D'4.ln. C.'o~a~ r.nL~lb CR5F~~ OOEirl 0E31 'O
LI'I~ EP.":LI LI-rLDO 0Ili~. 0nlp-J2 0~5Fs C~5029 0PTL23 OPlrPI C.OC~n7 03~'-'(
Cln~~ IC"O~C 0"72P5 0~l.l~ 0~1r1~7 COlrPL* O.q~1~0 07~17C~ 0"212C 0; 3fl ~Z
wopg C3reP' II`Cn?0 OC\-?i 0'c~l~~ o.o.ro, 0.-rrr~p 0~'llT C97.,50 n.oPldh Ca~r~:J
r?13 C~Ln37 0Clo29 I1'C^) nOH~FI 11."5CIO 0OJ'133 0"R113 090"02 CCbyOl OCJ'~'
U".11 0041'~ 0CL~J9 P~I~LI 1011)(1? llC10~11 0.1570h CO~FLC 370.C. F02707 OCO: IT,
*"~c C"h-'L D~~(~1 0~L)19 nol~rri I.CO*-n ~.112o* C'1LoO 0~11. ~sjc,. ralllE.
rnc C.rr7ci C40~3C
o.'iEOZ 0.11:11 O."~'pr U.S~%'~ l.~i00~ Oc~i92 070C~O ~Ll~b
C""7rn CCr115 ~E'"lll 0a"9rg O'~IIP 01`"9nl ICOCO O~Ccl" 010Q~7 C~71'3
1'Ur, C71L'C 1.4'090 O.~~-~p 0.7rnCq ~ 1.1C50 0.7CL1O l.bCn00 0"~2~2 $~li~O
O'r~l E."J'Z' OcZlsF O."CUI n.~2,0~ r.nr.c72 0~d 17+ 0"OP7 0.90??2 I.~C010 Or~J7P
11Unp P1*13? 0r~cq3 0E.1'~-7 0.ES1IE. OnC7.lo Oalal~ 0"'l.o~ Ohl..~ 0~F17P ICOrrO
IVn~ 0~LICI 0~71'9 0~'51L O"b.lPr. P~TIC~~ I Omrao F.r~l,.'a OCEncn CCCEE. C7P L
'~V"~ P.'icle E~'~75 0a2115 OOIITP rP4qle O. os~~O r. o(l 7'0 0.1Jng7 C.o3is7 OE1'^3
!ul C.';"PP n.,rt0s o.-lI07 r.e7nne CL"'~I C...LI.I O.~~E~, CYSL'R
rui2 C~"Tto O'")CO 37'l~i n7n0aa Ct~lPI 0'29~0 O.?nr.2 0.,~9~0 O."IJa* 5rl"';
CI"Pa P~ldnl O"rn~Z I).OL~nl~ 0 ""'a 011-17 0'~'02 OI1J'EL. 00141j g..s77 CrE'25
LlnlCL ,00.sr*r -O.l'sPn -n.~srqR -P4~~10 01071(9 -o.l~c~y Olsrc7
Y41T9 -Obllnr -0.1"~)I -OIJbTL~ -0.60'.17 -C~L.!~R -OUlOlo -COorl~ ~.7Lc~~
*r!C- -0.(l]"q --0~,10r O.nrrn? -OL.I'.'~ 0.*Il--C
""'I-~ ~n.ri~(i ~-C~'n~~ -C.7(r. I -O~71~o -C.7iYIO -Od~OST ~Cll'rl
"'i"CII ~C5~~L' I -I4rLnl _CL~Inl~ -Ohl4hc, -Cc*Oy? -0ICC31 -o air r( ~1OL 7 ;j
'ICUCI 1 'E.cP ~07171r ~O.nl~cn ~".'L '19 -C O"' 59 -014C03 -0C7'"'
'Or CP712i PPZPOO IJC.C. j..lJ*"3 0.(.'4LO 0J153~ 0J0.101 r~crl. -CC7~~~9
r!rr -0'~n0~ -o.~r-o? -0.-7 -c.'r~; .*. ~-~ If o -c. ~.I~ s -u.or~7 -OC~II' -I ~JY*T
v~u C~69~L -O1CCI~ -~~~147 ~r~C417 ~CP40n~ -0hd70~ -0.b5ari -(.~CCLI
ur:r -~ib07qn -C6~C35 -1)3'072 -n~bdl? ~0~"B'~ -CJ'.bllO ~C~751~ -03P~Z
Sqt COI~~C --CIC130 -]01 rs -c.rie~~ c.o~112 o. 0~9~ o.oln~~ -075'70 -" LCOLO
-E~0n7L -02~C49 -01na41 --C~"l'~ -C~0h~7 -0.21097 --I.IC100 -P.ln~7~ -n.l~L3f
'5* 0OP3E3 0O"~L9 ~0'~23 n.0340* -tCZf72 0.01007 0O~J*7 005~78 1107C33 -0 1O2'1
03 -O+EL929 -01~190 00?~dZ -0OL300 -0IF~~~ -Clnnll -OO?~Oa -010010 C.Ca~Z4 002~'T
luO+D -O.vIo.lI --01"~7 --0232~5 --0e~"0 -Cpn~rh -0~~957 --0E59r7 -0130034 -0 'co1l O.~~u^e
1uOT4 -0I~'OI -01P139 -Cl~'ns --01~~7~ --0. 17014 -0.11193 -e.Lnl~~ --CI~OPI -0.ioh13 e21L~1
!UDTS -039~5~ -OcCnSp --JZh~OT -0.2"'21 -Ol"gor -OP~P~~ -0.203011 -0.424rh -O.aCZI~ 0.191"7

































OI'1L

S.q:ILZI
OLI1'~
J7"cL9
C79~1a7
0on*~r.
Olr.Eb~'
0l(,c~
I?"?i~

C.?~-r-
T~Sbr.l
CO+T-r~
1. o~l( 7"
C"~LZI
C.'~13C


n7~~(6
CL~1ICL
G~~rO`
17'1'1
n.171'1
Csrcll
CPb7"4
r~7CEal
O~~a~g^
C*IP'~'

O.1FCs~
O.~~OL~
I~10C^

o.~-r~r
C1EL(L~

On~la:
C"llr~
O.Cllon
~012"
~0?IE""
-O.azazn
0. las?~r
-rL~0?4
-eFII1~%
-OCdPdO

~Clol~e
oOOgs"
-O.CI3G~
-0I~~P1
-0OP~5
-0ZRL~9


CluP5

0PZ35h
e.~acz
OIClrip
0~2357
n.P7ChO
0+6005
a.L~cP90
CrrZZ~
Onlf EJ
0Cb752
O~'%~C"
CCIPZI
0Cc2)2
0~1C07
07(-711
0P1134
n.F702C
0~L211P
0~"rp~
OI"IL'1~
O.rrLlq
I)'c77e
O~ZbCZ
n.cllZe
0.CI~I~
O.rr~rl
Ccrp~i
OCi~~L~7
rCJ"'
'J'~~3
?.C17ng
I 1Cc~O
F"Pi~h
O.~c7C2
0,'~IJ
-0IC17C

-n.iric
-o. I~IR
-O~nr7a

OIFP~n
-C7101"
-0'-PR07
-O~sp~r
-CZCC51)
-0~~r'lS
-~:ICZO

--J3LbeP
-O.polBr
-041eOe


Clul)l

011215
0L3172
041'42
OLr~ER
0`01~17
U91 )jl
0.,.)~70
nc2)19
o.~si~o
0'21'15
0~2' L.1
0~4'24
0'5*Iq
CP~717
0. IZ'5~5
C"l'ir~~
OE~`~'iq
012:')a
07~1'4
n.''r~L7
0.'~-i09
r 11:75
O~JLP
"'~ICP
Ci~~R5

C~17'~1
O.~~e~O
0~I~~~
L)ni~i(l
n7:'-C
0.,:'3s
I~]CiJ
O~O~rp
Ocn~n

-I].`C"17
0 10?J1
-0li9"11
-0. 11~14
-0~slrnq
D. Illo,
-0'I (48
O~Li'lr)
-".GI~**
-0.n2117
-0"I'I(-
- O. 8';n 3
-0~7C"e
-0.2511
-0 lbc15
--0~4515


CIUDI CluDa H~-ltl HIICII

9FE04R P.el~l' 0111317 -CJ7042
0~~5,4 Cr0'57 CIIIR~ -001157
0'1"~2~ 0r~7~~ -017756 -060130
ETFrn ~P01517 -P~PF07
OF~~~n 0.'1CI ?1~221 OOZ"~r
07107a P. c0:isq -C02LIO -0C~ZC7
0."5517 r.~23~0 -0111~0 -0.cn3rq
07~2~3 6-1'~~ C00008 OOOU'O
0n7L3+ C`1171 C~~70~ 005702
011~~0 n~Erna -11652~ -0U9n19
O7'boC 0"pc17 ~CL~175 -t~09~Z
07L.1CO CC~~~O -a06L93 -507.(11
F.FoJC~ P.r0110 n.1~51~ GllhcZ
OTnqc% d'4''.1 --D.1~17d5 -Cl~~t7
CdT'I~J o.77111q -.C7.697 -O~u~il
CCESni C?lan~ -0.2~129 -C.~17r4
~LCPFb P.Y7r~o 'r1~710 -C.llb.h
O~lldi CFIFI1 -0~n~5h -CL2011
0'4~~" r10371 -0.11111. -OL~L.J7
O.~~qrr CFCi"Y -04~"03 -OL.IIC~
0.9CZln T."ILJI -0r4~95 -O.~N~U
lcrJC~ O"1Y'Z --O. 30 755
0'712~ CI.*hCI, --0..i0~74 -C.19aC~
07COd~ n.F-~ir -0~9~19 --C3P.%C
Oh~lZI (75r~7 -0.415L5 -G.~1C7
".52~70 0.70"*.' -C.F~~~q -561~79
0~'hr7 *. aJ~s,. -CnJ910 -0.3~3n0
0774P~ 3.C3CIJ 0107109 -OPlOlo
OUIICO O~Or I -P1151J -CP~~C~F
Orl*"? O.rZ"" 0l~~n7 07~0JJ
U.~Trlc; ?.;2~'rn 01129~ O.nlln3
O.~~~re O.r7r7~ -019~~~ -0L7~31
O.Lnorg 0;14a -C.O"C~7
I.PCOCO t"~rlO 0.1971) o.OCc27
0P~410 I05C'1 -O.CILOL -0.0'15"1)
C131'0 1.00030 OL.nch
0."077' Cn:?q~ I.OCnOO
~I~:~e n.l~C.74 C'1D47
-OPd~O? P911.1 0 ~ l.'d 9
-n..ln17~ -C297~.h n57973 07,"91
-0~l~~q? -C. E1~.17 rnCPL Oc~~:lh
C. I~lcO 01~"" I CP~"IT. -o.il~h-~
-0.;4-155 -Cro, ~.r
-CZC~~7 -CC)~ Z ~0~5h~4 -00~2"7
--Olio~R --C"Oa.n -0~220~ ~~1~00
--O.I~'"U -C.~.077 --Orni77
-Olrs~r) -C.al7~~r C.0~~,5 Ori-l~
-0075~J --Q~5~oS 021~55
OI~1Ca -C2171h ~.lnT'Z 0.301110
O1IC(i79 -o.?nb~c 0~92~7 OILP723
00400~ -0~0151 -00997" O~ctl~
~0PZOIZ 0~I 9.91 OL1ICJ


HblCI ~4HCb H5~CO

0,05513 -0,357.10 -0.3chr4
-0O~L41 -02019r -C3Cll~
-O.PlnlO -o.ls(EZ
-065311 --:.ZO"'I -C~~:,5
0.11"02 --r.E~~C0 -3,02LBl
-0OZC~3 -C~ca~r
-0.10514 -5.11cLe -"~21r9
O I)OIF2 -CL~OO( -E l~nr7
0,13157 -0,52-75
-o.nira -OTPLn~ -0,'lrrb
--O.LrC~9~ -nLI?~IC _C~^C.C
-0nrrp~ _3SblhP -r.7~L31
OI~"Ch -~COFI p~Cir~C
-allF71 -C.C7~OI -C~~~5o
-0~C105 --IhllZL --C~C!I~
-0125073 -^n7*~1 --3C"cP
-OIrleF
-0~L~33 -Cr729a -IZI'GU
-03~LE~ _OL~C~C -~C2iil
-C.aF~~o -lrn3?7 -r~~19C
-05r;'0 -C7gc"L -OFCcC~
-0~"1CI -CC"L4~ -~L7'11
-OJI~Fc --1.~.7.11~ -)..c7a.
-n.~~iOf ~D~20'7 -0""CF2
--n41~61 -n7~r14 -OLPLdO
--e5~PY~ -7.Flc~~
-0,46753 -1.7C~IO --C.F'712
O~((~~ -Ialri7 -OIPa~i
-01~~24 -~C1"~1 --O.~?CI
0E100~ -C17CL~ -001'F~
013199 -Cb~2qE -321iP5
-0Icl~l -C'LCIF -7E.'71
o.anl~i -?43n)l ~n~1~51~
o.i~z~z _n,76C*o -O ~LI~~
-0;2420 -h.CC129 --0.2P7ct
001F~I CP~IFL 0'7~73
C53"LE 5~1Cac C.~3cc~
ICSI)UO 0C.3"qr. OE)7LC
oi~~P~ IOCPOC C143L3"
0577R0 C.43r3' LCrlT~
0,cp*qh C7q194 0~0407
01'15j -0.?n..c0
o2COIP I~7rCO 0.1~"74
-n 3"157 nlpsra -C.1~5E~
a7~103 OLC45r 0CP^~7
-0~~07~ 010"a~ -0.10120
CIOLIR 0,44440 0312nl
-021no0 O~Oole C.ll*h5
oEI~Ao ~E~015 007151
031~~4 OOLlr~ 0~5"~
-0,J57rr C21~F0 OJe;~
50~137 -C07~5 0llL7~


UD~~
uo~l
uong
uog~
G"5
G'L
TEI
GSd
ri+F,
~i3L
CI~P
d5~
Ylb

unli?
U*ll
unrq
rnl

u~~
un~r
Imis
k019
Y"~I(

u-e
uDar
1!'115
OIu~l
9U1?
I!uL
C!'15
r~11
CIUII2
~~ul)L
Yb!r~
*I!rs
IlalC~

*EJ(CI(
L15*-T
!"1
C'"

UFT
TC-
'TJ

1~
Ir~r~
lupm
1UDTr


TABLE 4 CONTINUED






















TABLE 4 CONTINUED




I5'1B Tol CID V4H YoT SST PRJ FS* g) rrcrp


upgI -0.39616 0.11444 -0. Wil? --0.47372 -0.4rgir -c. 111P6 -0.aa440 -0217n 0.042635 -O.42113
-0.2r,33 n~c ~ -0a5.0127s -n.4@3sa -d.Tears -C.3stra 0o.00012 -0.30204 -0.207-6
""'2 -0.51402 0.26F316 -0.767.14 -0.Fei39 --0.63Ing-c\77 -0.26747 -0.10004 -O0.2tla -0.12530
ur,?1 -0.10470 0.24422~ -0,at11 -0. &303~ -0. 3OI -0.r93a -0134L6 -0.0aA37 -0.~0200 -O.400 a
5C2 -04^ .20254 -0.44721 -Onlanac -0.30222 -0.11285 -0.111P70 007C12 0.00 21E -c.I124A
5'1 -0.43010b 0.0529S -0,aest6 -0.r7122 -P.e35crS -0,40033 -0Co309 -0.74925 -0.RI o -0.10 me
n"I -.471 -0.007Or -01a5 -0.72430 -0,5201r -0.17 14 -0041R -0.23164Z -0,qo 0 -0.230 7

001 -CAl 0.022 n -55o -0,FRT- a -.s"271 -c.capon -o.~sago -0179 .11131 -0.0rzTA -c.402 =
',"2 -C.6"CL7 0.Er?5-maan -0.Ja201 -0.~32220 -C.07192 -0.22360I 0.11109 -some 5l -0.C3 4~
6n91 -Opens.0 0.22/0 -O. 0? -Delotto -C.43^35 -0."Al30 -0.PREAR O.00240 -0.12150 -0.Alac'
"2l -C.b7207 Opent 3. T -n.2=95a -0.Tats" -0.04505 --0.19170 OL7P33 C.Opea -0.C232t
403< *C.7773. 0.Tpo3 1. -P.245'~3 -0.10672 --0.01244 -C7n"Co O .03& -0.1E77C --Cpiner
war -0,~10"" o.30'no3-.T0 -o.24005 -0.5*'440 0.01435 --0.240<7 0.02T~37 -0.peaLo* --c.;sr:.
05<1 -0."11 43 0I)Pl -0,041 -071rda -n.57~iT -0.37551 -0.20104 -0.05474 -0.1010 -0.30054
OTUDI -0.51 -FT 07 O.fCO -.(1177 --0.A'1Al -0.39 21 -0.25e1aa --0.'127 -0.0531' (05a'(r -Oage 4
1?"02 -r.1104 *C.C0101 -0.34 82 -0.40 ALr -C.1464 -d. 7060 -C0,tmathL07 -0.19277) -3P01C,2o e
O~uoA -0.42420 0.r51s0P -0.6413 -0.07321 -o.nEan -t400 -0,~1:o9 -0.002< -0a000' -0.170 ?
U?'9 -C.AS'ls30 ca aGL- -0.'tiis -O.52 07 -0.C4023 --0.2n05A -0,33Loa -0.01520 -0.1 #25 -a.34402
Unt 31 .0.- 74 0.11309I -0.*t36 -.'laPO0 -0.61700 --~0.011 -0.PARR7 -0.14-43 -0C.4 ^0 -0.2ast?

AT4 .40413 0.120, IB et 84 0.3A -O.3270 1 *C1Ao7 .05 --0~~4 0.2574 -0.1-072 0.74277
SCOL C.20I34 *0.4164 01 ^ -0).0325'a 0.3A0 -0,400 -I0,472 0.Oa2435 -.217"? --,d27PT

1969 -0.racy 0pe50 012' -0.13ollo 0,00077 -.~~ o -0,Pu?0 0.19 0.11409 -0.073"3 0n.207=
oT1 -0.37140 leCrZEe ia:1 -0.1742" --0.70s"2 -0.r'73 -0t.17rP9 C0.1715J -.01724 --0.029


Ysn 0200 0.110..i 1 41 A .i4090= r-as 1- 0,4 $'94 -o.=1544 1.00000P 0.703071 -0.35cc3
OJV~ 0.446'2: -0.01'24 0.65.,11 I.T 001 0r. 30020 0.20747 0.l0301 01.00000 -0.03ce :
IOTOli 0.097 -prci .32FAT 0.1748 ~ 0.20s9oS -0.tas?3 -0.ln7p5 -.70*3ca -o~ ~s 2gc.4




















TABLE 4 CONTINUED





In Ta TwstP





I"." ~ a. cii-- (201












~~F.17 -(012J'
0.3, a- page

c-1 ~ gpe1 -** 7n7r-i.' I?


-'? 3 .pgret

-0..014r -c.3'tL1

c~l -. Imp mer?

vt7~ 0..3)0 l` ef 0











TABLE 4 CONTINUED

LEGEND OF ABBRZEVIATIONS USED FOR
ELECTPROACOUSTIC AND BERAVIIORAL MEASURES


MlPOS
MP01
MrPO2
MP~OA


Maximum Power
Maximum Power
Maximum Power
Maximum Power


Output at .5 K~z
Output at 1 K~z
Output at 2 Kliz
Output Average of .5,1,2 KHz


Gain
Gain
Gain
Gain
Gain
Gain
Gain
Gain


with
with
with
with
with
with
with
with


Input
Input
Input
Input
Input
Input
Input
Input


at *S K~-z
at 1 KHz
at 2 KC~z
Average of
at *5 KF~z
at 1 KMz
at 2 KHz
Average of


*5,1,2 KHz




*5,1,2 K~z


HD4 .
HD8
HiD12
HD16
HDAA
HD1
HID2
HD3
HD4A
EDAB
HD15
HD3K
HiD45
HD6
HDAC
QIIMD5
QIMD1
QIMD2


Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
Harmonic
harmonic


Distortion
Distortion
Distortion
Distortion
Distortion
Distortion
Distortion
Distortion
Distortion
Dis tortion
Distortion
Distortion
Distortion
Distortion
Distortion


at .4 KHEz with .4 KHz Input
at .8 KHz with .4 KHz Input
at 1.2 KHE with .4 KE~z Input
at 1.6 Ktiz with .4 K~iz Input
Average of .4,.8,1.2,1.6 KHz
at 1 KHz with 1 KHz Input
at 2 KHz with 1 K~z Input
at 3 K~z with 1 K~z Input
at 4 KHE with 1 KHz Input
Average of 1,2.3,4 KHz
at 1*5 K~z with 1.5 K~z Input
at 3 K~z with 1.5 KHz Input
at 4.5 K~iz with 1.5 KHz Input
at 6 KHz with 1.5 KHz Input
Average of 1.5,3.4.5,6 KHz


Quadratic
Quadratic
Quadratic


Intermodulation Distortion at .5 KHz
Intermodulation Distortion at 1 KHz
Intermodulation Distortion a't 2 K~-z











TABLE 4 CONTINUED




QLMDA Quadratic Intetrmrodulation Distortion Average
of *5,1,2 K~z
CIMDS Cubic Intermodulation Distortion at *S K~z
CIMD1 Cubic Intermodulation Distortion at 1 KHE
CWID2 Cubic Intermodulation Distortion at 2 KFz
CIMADA Cubic Intermodulation Distortion Average
of *Si,1,2 KHz
HAICA H.A.I.C. Bandwidth Above 1 KEIz
HAICB H.A.I.C. Bandwidth Below 1 KHz
HAICT H.A.I.C. Bandwidth Total
HSHCA H.S.H.C. Bandwidth Above 1 K1z
HSHCB H.S.H.C. Bandwidth Below 1 Kf~z
HSHOT H.S.H.C. Bandwidth Total
IRI Index of Response Irregularity
CID C.I.D. W-22 Word List
VAH Veterans Administration Discrimination Test
MRT Miodified Rhyme Test
SST a synthetic sentence test
ESJ C.I.D. Everyday Sentencess Judged Intelligibility
ESW C.I.D. Everyday Sentencess Write Down Response
QJ Quality Judgments
IMDTO Intermodulation Distortion Test S/N=0 dB
IMDT4 Intermodulation Distortion Test S/N=-4 dB
IMDT8 Intermodulation Distortion Test S/N=-8 dB
CIDRT C.I.D. W-22 Word List Retest
VAHRT Veterans Administration Discrimination Test Retest
MRTRT Modified Rhyme Test Retest
SSTRT a synthetic sentence test retest
ESJRT C.I.D. Everyday Sentences: Judged Intelligibility
Retest
ESWRT C.I.D. Everyday Sentences: Write Down Response
Retest










TABLE~ 4 CONTINUED

QJRT Quality Judgments Retest
IMDSTORT Intermodulation Distortion Test S/N=0 dB Restest
IMDT4RT Intermodulation Distortion Test S/N=-4 dB Retest
IMDT8RT Intermodulation Distortion Test S/N=-8 dB Retest











TAELS 5



SUMMdARY OF THE FOUR HIGHiEST PEARSON CORRELA4TION
COEFFICIENTS BETWEEN EACHi BEHAVIIORAL MEASURE
AND~ THEi ELECTROACOUSTPIC CHARACTERISTICS


Behavioral
Measure
CID


Characteristics

E~FD12
HD16
ED~AA


r`

-.909
-.909
-.894
-.885


-.788
-.742
-*?32
-.728


-*755
-.756
-.683
-.632


-*553
-*546
-.509
-*505


-.566
-,542
-.508
-.4193


GFS
EDAA
MIPo5
ED12


MiPo5


SST


GPS
ED#
QIMiD2
HDAA


CJIMD1
ED~r12
HD1l6
HD8


ESJ









TABLE 5 CONTINUED


Characteristics


Eehavioral
Measure

ESW


r`


-*412
+.304
-.253
-.248


-.596
-.544
+*530
-*516


-.526
-*514
-.510
-.436


+.363
-*362
-*345
-.336


-.630
-.624
-.623
-.510


IRI
HSHCA
HAICA
HD8


HDi2
HD16
HSHCA
HD8


MPO2
MPO1
MPOA
HD3


HSHOB
MPOS
CIMD1 I
IRI


MPO2
MP01
MPOA
MPOS


IMDT1


IMIDTI


IMDT8


r = Pearson Correlation Coefficient







variable dimension. For example, all measures of maximum

power output (MPOS, MP01, MPO2, M/POA) in a hearing aid were

predicted to be highly intercorrelated because they all

measure dimensions associated with maximum power output.

Further inspection of the matrix reveals that many

characteristics not naturally expected to be correlated have,

in fact, high correlations. These characteristics cannot be

assumed to be independent of the characteristics with which

they have a high relationship, and to a greater or lesser

extent, are a measure of the same source of variability. For

example, MPo5 was not only highly correlated with other

measures of maximum power output but also with measures of

cubic and auadratic intermodulation distortion, harmonic

distortion and gain. Changes, therefore, in intermodulation

distortion, harmonic distortion and gain will1 affect measures

of maximum power output and vice versa. This dependence

indicates that there was an interaction between characteristics.

An electroacoustic measurement, therefore, is never a totally

independent variable, but the sum of the moment-to-moment

interactions with many of the other characteristics.

Reference to solitary electroacoustic characteristics in

the literature is probably not accurate. It is possibly more

descriptive to refer to interaction effects and define the

major components in the interaction. Not only would such a

description more closely approximate what is actually

occurring between characteristics, but permits a multi-

dimensional appraisal of the interactions affecting speech

intelligibility.







What is the Relationship Between Behavioral Measures?

Table 4 reveals that, in general, the correlation

coefficients between behavioral indices were low. This is

interpreted to suggest that each measure is affected by

essentially different sources of variance. Several exceptions

to this generalization are notables CID correlates +.83 with

VAHI; VA~H correlates +.76 with MRT; ESJ correlates +.81 with

ESW and +.86 with QJr ESW correlates +.70 writh QJ; IMdDT1

correlates +.72 with IMDT8. The correlations indicate that

these measures are not totally insensitive to the same

sources of variance, and to some extent ar'e redundant. The

degree of independence of the behavioral measures, however,

is remarkable, because these measures were all designed to

estimate and quantify speech intelligibility. Tecreain

of Table 4 signify that either all of the measures are not

estimating speech intelligibility or that the measures are

quantifications of essentially different and independent

dimensions of speech intelligibility.

H~ow are the Scores on the Behavioral Mieasures Influenced by
the Electroacoustic Characteristics?

To attempt to answer this question a multiple regression

analysis was performed between the electroacoustic character-
istics and the behavioral measures.

Statistical approach. AL multiple regression analysis

generates a linear combination of independent variables which

will correlate as highly as possible with a dependent variable

and produces a regression equation. The regression equation
can be used to predict the value of the dependent variable from







a set of independent variables.

The normalized beta weights, associated with the

regression equation, can be used to reflect the strength and
direction of the relationship between each independent

variable and the dependent variable. The larger the beta

weight, the stronger the influence on the dependent variable.

The multiple correlation coefficient (R) can be squared
to reveal how much of the variance of the dependent variable

can be explained by the prediction equation.

The multiple regression procedure was used in this study

to produce a prediction equation for each behavioral score

from a set of electroacoustic characteristics, provide infor-

mation as to the relative strength of effect of each character-

istic on behavioral scores and to calculate the percent of

variance in behavioral scores which can be accounted for by a

group of electroacoustic characteristics.

Special considerations on the use of multiple regression

analysis. Interpretation of the multiple correlation coeffi-
cient and predictions from the regression equations can be

misleading if certain assumptions for use of the analysis are
violated.

The first assumption stems from the fact that the multiple

regression coefficient is dependent upon the subject-to-

independent variable ratio. According to Nunnally (1963) an

unbiased R can be expected when the ratio approaches 13 to 1.

Lesser ratios have an increasingly larger chance of generating

a R which is biased upward. The number of independent variables







in this study was forty-two and the number of subjects

equalled twenty. This ratio would be inappropriate for a

multiple regression analysis because the R would always,

misleadingly, approach unity (McNemar, 1969).

Since it is often not possible, in behavioral research,

to have many subjects relative to the number of independent

variables, two procedures can be used which reduce the up-

ward bias on the Ri. One procedure would be to employ a data

reduction technique to minimize the number of independent

variables. Such a technique would improve the subject-to-

independent variable ratio. An obliquely rotated factor

analysis is a reduction procedure and was employed in this

study. A second procedure would be to use a formula which
would estimate the reduction in the R due to a less than

ideal subject-to-independent ratio. The technique is known

as the correction for shrinkage and was applied to all of the

Rs generated by the multiple regression analysis.

Factor analysis. Given the correlation coefficients for

a set of variables, factor analytic techniques can be used

to lessen the number of variables by grouping them on the

basis of patterns of interactions. The smaller number of

grouped variables or factors can be taken as source variables

accounting for interrelations in the data (Nie et al., 1970).

In other words, the factors are extracted from a large set

of independent variables on the basis of common patterns of

variability. The variables placed into each factor grouping,

therefore, covary with each other and are not independent

variables. The factor groupings, however, are as independent







as possible given a particular set of data. The end result

of a factor analysis is a set of factors and factor loadings

which indicate the relative importance of each independent

variable to the factor grouping. For example, given two

measurements of harmonic distortion and three of gain, factor

one might have high loadings on harmonic distortion and low

loadings on gain. The factor would be referred to as one

reflecting harmonic distortion effects. The name given to the

factor grouping is arbitrary and is determined by the investi-

gator upon inspection of the factor loadings on the independent
variables.

Rotation of factors is a technique which manipulates the

factors so that a variable (or highly inter-correlated group

of variables) loads very high on one factor, but approximates
zero on the other extracted factors.

The rotation can be performed assuming little or no

correlation between variables orthogonall) or assuming that

some dimensions of the variables are correlated (oblique).

Oblique rotations are commonly accepted as more accurately

representing the clustering of variables because few variables

are completely orthogonal.

An obliquely rotated factor analysis was performed on the

forty-two electroacoustic characteristics. Table 6 shows the

extracted factor patterns. Five factors were determined from

the forty-two characteristics. Inspection of the factor

loadings for each factor reveals that GP2 loads highest on

factor one; HSHCT loads highest on factor two; MPO1 loads

highest on factor three; HD12 loads highest on factor four and









TABLE 6


FACTOR LOADINGS FOR EACH OF THE FIVE FACTORS
ON THE ELECTROACOUSTIC CHARACTERISTICS

Factor Factor Factor Factor Factor
1 2 3 4 5

MPo5 +0.031 -0.017 +0.263 -0.088 -0.231
MP01 +0.003 -0.105 +0.442 -0.202 -0.018
MiPO2 +0.153 -0.020 +0.426 +0.019 +0.105
MPOA +0.068 -0.050 +0.400 -0.092 -0.044
GP5 +o.058 -0.034 -0.024 -0.248 -0.122
GP1 +0.216 +0.007 +0.021 -0.040 +0.002
GF2 +0.327 +0,047 +0.010 +0.129 +0.105
GPrA +0.204 +0.005 +0.001 -0.066 -0.011
Gs5 +0.073 +0.048 -o.144 -0.171 -0.135
GS1 +0.155 +0.065 -0.005 -0.028 -0.077
GS2 +0.220 +0.081 +0.090 +0.081 +0.003
GSA +0.151 +0.066 -0.025 -0.046 -oo?5
HD# +0.150 -0.027 -0.134 -0.185 -0.059
HD8 -0.008 +0.020 +0.026 -0.330 +0.061
HD12 -0.069 +0.006 +0.125 -o*383 +0.093
HD16 -0.038 +0.021 +0.083 -0.353 +0.088
HDAA +0.015 +0.006 +0.017 -0*321 +0.043
HD1 +0.164 +0.052 +0.035 -0.014 -0.084
HD2 +0.143 +0.155 +0.017 +0.004 +0.076
HD9 +0.048 +0.185 +0.076 -0.037 +0.047
HD4A +0.000 +0.221 +0.019 -0.047 +0.001
HIDAB +0.091 +0.159 +0.041 -0.025 +0.014
HD15 +0.132 +0.212 -0.088 +0.087 -0.032
HD3K +0.080 +0.228 -0.037 +0.060 -0.045
H-D45 +0.011 +0.265 +0. ook +0.028 -0.013g
HD6 +0.020 +0.258 -0.025 -0.011 +0.059
HDAC +0.061 +0.243 -0.038 +0.042 -0.008
QIMDS +0.158 -0.021 +0.020 -0.123 -0.065









TABLE 6 CONTINUED


Factor Pactor Factor Factor Factor
1 2 3 4 5
QIMD1 +0.191 +0.095 +0.006 +0.029 -0.004
QIMD2 +0.312 -0.089 -0.046 +0.029 +0.073
qIM~DA +0.227 +0.004 -0.000 -0.035 -0.011
GIMDS +0.110. +0.102 -0.029 -0.077 -0.036
CIMD1l -0.002 +0.012 -0.028 -0.275 -0.087
CIMD2 +0.296 +0.026 +0.007 +0.169 -0.035
CIMDA +0.125 +0.051 -0.021 -0.097 -0.060
HAICA +0.100 -0.281 +0.033 +0.060 -0.424
HAICB +0.263 -0.211 -0.431 -0.073 +0.153
HAICT +0.123 -0.288 -0.013 +0.049 -0.387
HSHCA +0.257 -0.299 +0.198 +0.227 +0.094
HSHOB +0.072 -0.280 -0.259 -0.215 -0.004
HSHCT +0.225 -0.334 +0.079 +0.112 +0.063







IRI loads highest on factor five (see Table 4). Other high

factor loadings can be discerned within each factor. Since

the purpose of the analysis was to reduce the number of

independent variables for input into a multiple regression

analysis, and because with twenty subjects a subject-to-

independent variable ratio of 4 to 1 was thought to be the

lowest possible for reliable and interpretable results, five

factors were the maximum that could be extracted. Selection

of the highest loadings for each factor, however, assures that

the most influential variables under each factor will be

included in the multiple regression.

The name given to each factor was determined by the

variable having the highest loading on the factor. Factor

one was termed a gain factor, two a bandwidth factor, three

a maximum power output factor, four a harmonic distortion

factor and five a regularity of frequency response factor.

Multiple regression analysis. The independent variables

derived through use of the factor analysis (GP2, ESHCT, MP01,

HD12, IRI) were multiply regressed against each of the be-

havioral measures (CID, VAM, MRST, SST, ESJ, ESW, QJ, IMDTO,

IMDT4, IMDT8). Table 7 shows the summary table of the analysis

for each behavioral measure. The electroacoustic character-

istics under each behavioral measure are arranged in order of

decreasing influence upon the measure.

Table 7 reveals that all of the multiple regression

coefficients are high (with the conspicuous exception of

IMDT4 where R = .50)). This suggests that the five selected

variables account for a large percentage of the variability









TABLE 7


MULTIPLE REGRESSION ANALYSIS OF EACH BEHAVIORAL MEASURE WITH
THE FIVE HIGHEST ACTOR LOADING ELECT~ROACOUSTIC CHARACTERISTICS


2
R R

*94 .88


2
Rs si
*91 .84


Measure


Characteristic


beta

-1.51
+ .44
- .30
- .25
+ .09


-1.93
-1.06
- .63
+ .58
+ .13


-1.28
+1.00
- .68
+ .26
+ .1


-2.47
-1.46
+1.38
- .66
+ .01


HD12
MP01
H-SHCT
IRI
GF2


HD12
:SHIC T
IRI
MP01
GP2


HD12
IRI
EKSHCT
MP01
GF2


HD12
HSHCT
MP01
IRI
GF2


HD12
G-F2
IRIl
M:PO
HlSHCT


.94 .89 .92 .85








.98 *96 .90 *95







.88 .78 .83 .69







.71 *SO *57 *32







TABLE 7 CONTINUED

Characteristic beta
HD12 -1.40
M1P01 + *73
IRI *68
GP2 +. .52
ESHOr;T .47


R R 2
*74~ *55


2
R s
.62 .39


Measure
ESW


HD12
GP2
MP01
IRI
HSHICT


MhPO1
HD12
tHSHCT
IRI
GP2


dP01
GP2
!SHOT
IRI
HD12


MPO1
EID12
HSHCT
IRI
GF'2


-1.39
+ 81
+ .37
- .13
-.08


-1.c3
+1.37
+ .69
+ *35
+ .05


-*4~3
+ .28
+ .279
-.21
+ .12


-1.66
+1.52
+ .68
+ .39
-. 07


.93 .87 .91 .82


INDTO


.69 .48 .54 .29








*So .25 .14 .02








.82 .67 *74 .55


INDT4


IMIDT8


normalized beta weights
multiple regression coefficient and its square
multiple regression coefficient corrected for
shrinkage and its square


beta
RR2
Rs, R







in each of the behavioral measures.

The squared multiple regression coefficient (R2) can be

interpreted as the amount of variance in each behavioral

measure which can be accounted for by the five characteristics.

In these data the range is 96P of the variance in MVRT to 259

of the variance in IMDT4 which can be accounted for by the

five characteristics

A second observation concerns the importance of each of

the five characteristics to the behavioral measures. The

normalized beta weights provide information as to the relative

importance of a characteristic to the overall R9. The

characteristics listed under each measure are arranged from

greatest to least importance as dictated by the value of the

normalized beta weights. The relative importance of each of

the five electroacoustic characteristics is different for

each behavioral measure. One consistency, however, through

CID, VAHi, MRT, SST, ESJ, ESW and QJ is that in these measures

harmonic distortion at 1.2 KHEi with a .4 KHE input has the

highest relative importance (highest beta weight). More

consistency appears in ranking important characteristics

between the Intermodulation Distortion Tests, but it must be

remembered that these tests are merely different signal-to-

noise ratios of the same stimulus material.

Correction for shrinkrage of R. As was mentioned earlier

a small subject-to-independent variable ratio can bias the R

upward in a misleading fashion. Estimation of the R with

reduced bias can be obtained using the following formulas

R2 = 1- (1 -R2) N -1
N- n










TABLE 8


PEARfSON CORRELATION COEFFICIENTS FORi EACH
BEHAVIORAL TEST AND RETEST
(QJ THROUGH IMDT8RT ARE GROUPED DATA)


Behavioral Measures r r2

CID with CIDRT .816 .666
VAH with VAHRT .871 *759
MRT with MRTRT .767 .588
SST with SSTRT .563 .317
ESJ with ESJRT .771 *594
ESW with ESWRT .650 .423
QJ with QJRT .827 .760
IMDTO with IMDTORT .981 .962
IMDT4 with IMDT4RT .768 *590
IMDT8 with IMDT8RT .473 .224







where

a2 = squared R corrected for shrinkage
R2 = obtained R

N = number of subjects
n = number of variables

Through the use of factor analysis the number of variables

was reduced to five. Table 7 reveals the shrunken Ri for each

obtained R in the multiple regression analysis. The shrunken

R is believed to be a better ie., less biased, estimate of the

R based on the subject-to-independent variable ratio in these

data.

The correction for shrinkage due to the subject-to-inde-

pendent variable ratio employed in the study appears as Rs and
R2 in Table 7. This procedure reduced to a greater or lesser

extent the R for each behavioral measure. The greatest

changes occurred in ESJ (.71 to *57), Esw (.74 to .62), IMIDTO

(.69 to *54) and IMDT4 (.50 to .14). These lower Rs because

of the less than optimal subject-to-independent variable ratio

are believed to be more realistic estimates of the effect of

the five characteristics upon each behavioral measure. The

squared shrunken R provides the same information as R2 and so

the percent of variance in each behavioral measure explained

by the five characteristics ranged from 95% in MjRT to 27. in
IM4DT4.

WEhat is ~the Coefficient of Reliability for Each of the
Behavioral Measures?

Pearson product-moment correlation coefficients are shown

in Table 8 for behavioral measure test-retest scores. The







squared coefficients are also included, which indicate the

percentage of the variance in one measure which can be

accounted for by the other.

Inspection of the table reveals that the test-retest
reliability for the majority of the measures employed in this

study was good. The synthetic sentence test (r = *S63); the

C.I.D. Everyday Sentence Test: write down response (r = .650)

and Intermodulation Distortion Test at signal-to-noise ratio

of -8 dB (r = .473) were the only behavioral measures having

conspicuously low reliability coefficients, ie., r(.70.
Cross Validation of Statistical Results

The data reduction procedure used to reduce the number

of independent variables in this study entailed selection of

only those variables having the highest loadings on each of

five factors. These five characteristics were, in turn, used

as the independent variables in multiple regression analysis.

The possibility exists that in another sample of subjects, the

highest factor loadings on each factor would be different and

thus the R and beta weights different. If any predictive

confidence is to be placed in the R and beta weights a cross

validation is usually required. The validation method usually

employed is to use the set of regression equations derived from

the first sample to calculated predicted scores on the second

sample. A correlation is calculated between the predicted

scores and actual scores, the strength of the correlation

providing information as to how stable the original regression

equations were and, therefore, the confidence to be placed in

the R and beta weights.







It was not possible to run this standard type of cross

validation. However, since there were available test and

retest scores on the same subjects a modified cross validation

was devised. If it could be shown that the Rs and beta weights

in the test and retest data maintained some degree of stability

in terms of relative value, then the Rs and beta weights

could be regarded as stable predictors of the importance of
the characteristics to the behavioral scores. A critical

assumption in this analysis is that the test and retest

scores for every behavioral measure are highly correlated. A

low correlation between test and retest scores would confound

the validation. With low reliability between scores, differ-

ences in the R and beta weights could be caused by the use of

characteristics chosen on the basis of chance high factor

loadings in the original sample, or produced by the instability

of the test measure.

Table 9 displays the multiple regression analysis of the

behavioral measure retest scores. As in Table 8 the electro-

acoustic characteristics are ordered in terms of decreasing

beta value for each behavioral measure. Multiple regression

coefficients and squared multiple regression coefficients are

also presented.

A quick comparison of the obtained values in Tables 8 and
9 leads to the conclusion that only the R of ESW (.74 to .90),

IMDTO (.69 to .59) and IMDTrC (*50 to .60) change appreciably

between test and retest.

The relative orderings of the characteristics on the

basis of beta weights for the test and retest data are presented









TABLE 9


MULTIPLE REGRESSION ANALYSIS OF THE BEHAVIORAL MEASURE RETEST
SCORES WITH THE FIVE HIGHrEST FACTOR LOADING
ELECTROACOUSTIC CHARACTERISTICS

Measure Characteristic beta RR2

CID HD12 -.94 .80 .64
GP2 +.16
HSHiCT -.05
MPf01 +.05
IRiI -.02

VAH ED12 -1.93 .96 .92
HSHCT -1.10
IRI .62
MP01 +t *55
GP2 + .15

MRT HD12 -1.50 .98 .96
IRI -1.03
HSHCT .69
MP01 + .46
GP2 + *456

SST HD12 -2.5 *78 .6
MP01 +t1.6
IRI .36
GF2 .08
HSH~CT .03

ESJ HD12 -1.13 .77 .60
GP2 + .60
IR~I .49
MIPO1 + .25
HSH-CT .22









TABIS 9 CONTINUED


Characteristic

HD12
MP01
IRI
GF2
HSHCT


HD12
MP01
OF2
IRI
HSiCT


MPO1
HD12
HSHCT
IRI
GP2


HD12
GF2
M~P01
HSHCT
IRI


Measure

ESW V


beta

-1.62
+ .89
- *S2
+ .48
-.18


-1.18
+ .68
+ .62
28
+ .26


-1.14
+1. 01
+ .48
+ .26
- .09


IMDTO


*59 *35


IMDT4


IMDT8


.83 .69


HSHOT
HiD12
GP2
N8P01
IRI


+ .72
+ .61
- .50
+ .29
-.15


beta = normalized beta weight
R?, R2= multiple regression coefficient and its square







in Table 10. Also included in parentheses is the Pearson

product-moment correlation between the test and retest scores

on each behavioral measure.

The relative rankings for VAH, MIRT, ESJ, QJ, IN:DTO are

identical for the test and retest. ESW showed one change in

ordering and the rest of the measures show multiple changes.

It should be noted that multiple shifts occurred in the

measures having relatively low reliability coefficients. The

notable exception is CID which, while the reliability coeffi-

cent was high, there were still multiple changes in the

relative ordering of characteristics.

The cross validation of the measures with high correla-

tion coefficients and none or one change in ordering m~ust be

assumed to be excellent, ie., the R and beta weights are

reliable predictors of relative importance of the five

characteristics to these behavioral measures. The cross

validation of the measure with low reliability coefficients

and multiple ordering changes must remain suspect because of

possible confounding introduced by the variability in the

behavioral measures themselves. CID while having a high

reliability coefficient should not be assumed to be cross

validated and, therefore, the R and beta weights are too un-

stable to use as reliable predictors of the relative effect

of the characteristics on CID scores.








TABLE 10


RANK ORDERINGS OF THE FIVES CHARACTERISTICS DERIVE~D FROM THE
ACTOR ANALYSIS FOR EACH BEHAVIORAL TEST AND RE~TEST


(.82)
CID CIDRT


(.87)
VAH VAHlRT


(.77)
MRT MRTRT


(*56)
SST SSTRT


HD12
GF2
MIP01
IRI
HSHCT



HD12
GP2
MP01
IRI
HSsuT



HD12
GF2
MP01
IRI
HISHCT


1 1 1 1 1 1
2 5 55 5 5
3.5 4 4 4 4 3
5 3 3 2 2 4
3.5 2 2 3 3 2


(.77) (.65)
ESJ ESJRT E:SW ESWRT
1 1 1 1
2 2 5 L
4 4 2 2
3 3 3 3
5 5 4 5


(83) (.98)
QJ QJRT IMDTO IMDTORT
1 1 2 2
2 3 5 5
3 2 1 1
Le 1 4 4
5 5 3 3


(.47)
IMDT8 ITMDT8RT
2 2
5 3


(.77)
IMDT4 IMDTCRT
5 1
2 2
1 3
41 5
3 It


(_ ) = reliability coefficient for each
behavioral measure









Discussion of Results

The validity of the reasons conjectured for the dis-

agreement in research results relating electroacoustic

characteristics of hearing aids and behavioral measures

designed to estimate and quantify speech can now be discussed
with reference to the results of this study.

Inadequate appraisal of electroacoustic characteristic

interaction has been suggested as a reason for the differences

in observed research results. The correlation matrix relating

the electroacoustic characteristics studied in this investi-

gation (Table 4) showed that there are patterns of interactions

among the characteristics so that no one characteristic is

ever an independent variable. Each characteristic is affected

by the moment-to-moment interaction with all of the other

characteristics. For example, maximum power output at 1 KHEz

(MP01) is highly inter-correlated (r =>.80) with GF5, GF1.

OFA, G-S1, OS2, G-SA, HDAA, H.D1, QIMDS, QIMlDA, CIMD5, CIMD4I1 and

CIMIIDA. At another moment with different input levels an~d

materials, battery voltage changes or component deterioration

the interactions and the relative strengths of the interactions

might be quite different. Smaldino (1972) found a similar
result in a correlation matrix prepared in a similar fashion.

The occurrence of moment-to-moment interaction between

electroacoustic characteristics reouires that investigators

perform cautious and systematic assessment of hearing aid

electroacoustics. Different procedures involving different

test stimuli, at different levels, will probably each create







a unique set of moment-to-moment interactions. Each set of

interactions probably affects speech intelligibility in

different ways and so result in contradictory conclusions.

Results from this study, therefore, tend to support the

conjecture that electroacoustic interactions have not been

adequately assessed and controlled in hearing aid studies
relating electroacoustic characteristics to speech intellig-

ibility. The obvious need is for a standardized set of

procedures and materials to assess the relationships between

intelligibility and electroacoustic characteristics.

The implication of the results is that standardized

behavioral measures, sensitive to differences in electro-

acoustic interactions should be employed in studies relating

these interactions to intelligibility. Conclusions, therefore,

on the relationship between interactions and intelligibility

would be based upon subject differences rather than differences

in the sensitivity of the behavioral measures. A standardized

measure would provide an index of subject performance which

could be duplicated in any clinic. The result, therefore,

would be recommendation of hearing aids which have been

assessed along the same parameters. This would provide a

systematic and consistent approach to evaluation procedures
rather than the inconsistencies in recommendation which

currently occur. For example, assessment of performance of a

particular subject on the same two hearing aids might be
reversed when two different behavioral measures are employed.

The results of the assessment would be used to recommend a

different hearing aid with each behavioral measure. While








perhaps an extreme example, the point is that the behavioral

measure can influence the outcome of the hearing aid selection

procedure, while our goal is to select the aid on the basis

of optimal subject performance alone. Use of standard

behavioral procedures would reduce the inconsistencies intro-

duced by the test procedure.

A further advantage of standard behavioral measures is

that a very thorough investigation of how electroacoustic

interactions affect the measure would be possible. With

concerted effort, the many complexities of interactions could

be traced and related to subject performance. This sort of

analysis is not available for any behavioral measure to date,

but obviously is the most critical assessment that can be

performed if the measure is to be used (as most are) to

influence the future design characteristics of hearing aids.

The final reason offered to account for the disagreement

in research results is poor test-retest reliability of the

behavioral measures employed. The different characteristics

that have been related to intelligibility could be a function

of the unreliability of the test instrument and, therefore,

not clearly reflect the relationship between electroacoustic

characteristics and intelligibility. The coefficients of

reliability for every measure used in this study was shown in

Table 8. It is clear that except for IMDT8, ESW and SST high

coefficients of reliability (r ).70) were found in these data.

The occurence of low coefficients, however, indicates that

they can occur in this type of research. Caution must be

exercised, in that reliability of a behavioral measure cannot








be assumed in studies relat;Ing electroacoustic characteristics

and intelligibility. Although not specifically investigated

here, the reliability of a behavioral measure could change
when used with different electroacoustic interactions. The

implication, therefore, is that a reliability coefficient
should be calculated for each behavioral measure used in a

study of this sort. Failure to quantify the reliability of
the measure can, at least, confound and, at worst, invalidate

the conclusions relating the measures to the interactions.

Audiologists are often called upon to recommend one

hearing aid over another, based upon evaluation procedures

which involve a client's response on behavioral measures

designed to estimate and quantify speech intelligibility.
It is clear that the behavioral measures studied in this

investigation were not affected in the same way by the five

electroacoustic characteristics. This implies that the

estimates of intelligibility derived from these measures

would be different between aids, not only because of subject

performance differences, but because the test itself inter-

acted differently with different electroacoustic interactions.

Estimates of a client's performance, therefore, would be

confounded by the test instrument. Since the audiologist

is concerned with the performance of the client when

recommending hearing aids, it behooves the clinician to use

behavioral measures which can be confidently interpreted as

indices of unconfounded performance.

One way of reducing the uncontrolled confounding of

estimates of performance would be to use a standardized test








measure. The requirements of such a measure would be that it

be affected by electroacoustic interaction differences and

that it be reliable. A number of behavioral measures

presumably meet these requirements. Most, however, have not

been analyzed in terms of what kinds of electroacoustic inter-

actions affect them and to what degree, nor has there been

any massive attempt to estimate the reliability of measures

passed through hearing aids.

The results of this study indicates that the Veterans

Administration Discrimination Test (VAH), the Modified Rhyme
Test (MRT), the C.I.D. Everyday Sentences: judged inteligi-'

ility test (ESJ:, quality judgments (QJ) and the Intermodula-

tion Distortion Test at a signal-to-noise ratio of 0 dB

(IMDTO) are sensitive to differences in electroacoustic

interactions, and are stable and reliable behavioral measures.

The order and strength of effect of each electroacoustic

characteristic were different (KID12 was most influential in

all but IMIDTO, see Table 11) in every measure. It might be

assumed, therefore, that each measure was sampling a different

combination of electroacoustic interactions.

A4 possible application of these data would be to use the

five behavioral measures as a test battery for use in hearing

aid selection procedures. Scores obtained on each measure by

a patient could be analyzed in terms of the electroacoustic

characteristics which have primary influence on each measure.

For example, a patient might be given the five tests through a

particular hearing aid, resulting in the following performance

scores: VAHi = 80,;, MRlT = 86j;, ESJ = 90 QJ = 70 ,i~ IMDITO = 68 ~.







For this particular patient and hearing aid the

inconsistent scores would be those derived from the quality

judgment material and the Intermodulation Distortion Test.
Reference to Table it, reveals that in order of strength of

effect harmonic distortion, gain, maximum power output,

regularity of the frequency response curve and bandwidth

combine to affect performance on QJ and maximum power output,

harmonic distortion, bandwidth, regularity of the frequency

response and gain affect IMDTO. Since harmonic distortion
is an influential variable on all measures but IMDTO, and

performance on VAH, MRT, and ESJ was higher than QJ and
IMDTO, harmonic distortion is probably not affecting the

patient's performance.
Gain is influential on QJ and ESJ, but performance on

ESJ is not very different than VAFI and MRT, so gain is

probably not most influential to this patient. In a likce
manner regularity of the frequency response and bandwidth

can be shown~ to not be variables influencing this patient's

performance. Maximum power output, however, is third in
influence on QJ and first on IMDTO. The same characteristic

is fourth on the other measures, and has relatively low

influence. Ai reasonable conjecture, therefore, would be that

the patient's performance using the particular aid was
influenced by maximum power output. The clinician might then

adjust the maximum power output and readminister the test

battery. For instance an increase in maximum power output

might increase QJ to 82;1 and IMDTO to 7934 leaving the other
scores about the same.







Another aid using the same subject might produce a

different pattern of results with a different characteristic

having primary influence. The key to selection, however,

depends upon whether the influential characteristic can be

adjusted to the point where scores on the measure sensitive

to the particular characteristic become consistent with the

other obtained scores.

The possibility exists, of course, that because of inter-

actions among charac-teristics a change in one will change how

the other characteristics affect the behavioral measures. In

this sense a method of minimal changes must be employed,

wherein, adjustments are made on characteristics so that other

characteristics are minimally affected.

Of course if low scores are obtained on all measures,

either the patient cannot effectively use amplification or

is sensitive to parameters of intelligibility not represented

by the five measures.

High scores on all of the measures indicates that no

presented combination of electroacoustic characteristics

adversely affects the patient's ability to perceive speech.

This strategy is hypothetical, but serves to suggest a

way in which information concerning the relationship between

electroacoustic interactions can be systematized and practically

used in hearing aid selection procedures. Evaluations of

patient performance on behavioral measures passed through

hearing aids are presently inconsistent and, may be invalid.

An outline for change is needed; what is described might serve

as a first step toward such a plan.









Implications to Further Research

The results of this investigation suggest some directions

that further research may take.

The statistical analysis could be confirmed and precision

increased with the following amendments. While the results

of this investigation were cross validated in a modified

fashion, a traditional cross validation employing another

sample of subjects is believed to be necessary before

absolute confidence can be placed in the predictive worth of

the multiple regression coefficients and beta weights.

Another, larger sample, should also be drawn and subjected to

the same test materials to reduce the chance of bias intro-

duced by a low subject-to-independent variable ratio. A ratio

of 10 to 1 is believed to be reasonable. While provisions

were made in the present study for a low ratio, it would be

of interest to test the estimates generated by the provisional

techniques against completely unbiased results. Finally, a

larger sample would permit inclusion of more independent

variables into the regression analyses. The factor technique

employed in this study may have capitalized upon chance high
factor loadings. Inclusion of some of the other variables

with high loadings could reduce these chance effects.

The degradation of the behavioral measures in this study

was speech spectrum noise at specific signal-to-noise ratios.

These ratios were arbitrarily selected to avoid an expected

ceiling effect of the better fidelity hearing aids. Other

ratios and other kinds of interference such as intellective







masking or environmental noises might be tried to approximate

various listening environments. Such information would be

important to determine if different electroacoustic inter-
actions affect intelligibility in different noise environ-

ments frequented by a hearing aid user.

It would be very important to perform this same type

of experiment with various sorts of hearing impaired

individuals. Jerger and Thelin (1968) asserted in a paper

comparing normal and hearing impaired individuals responses

to hearing aid amplified speech, that persons with flat

sensorineural losses approximated normal hearing responses,

however, a decreasing correlation existed with increasing

audiometric slope. From these results they concluded that

"one cannot generalize from behavioral results on normals to

behavioral results on all hearing impaired subjects" (Jerger

and Thelin, 1968 p. 175).

It is suggested here that Jerger and Thelin's statement

may, indeed, be the case and that the electroacoustic inter-

actions found to be important to normals may not have the

same relationships in various hearing impaired individuals.

The behavioral measures employed in this study were

chosen on the basis of accessibility and representation of

the major types of measures. The selection was in no way

inclusive of all of the behavioral measures employed in

hearing aid selection procedures. It would be important to

perform the analyses suggested in this study on the majority

of routinely employed measures. Statements as to sensitivity

to electroacoustic interaction and reliability could, then,




79


be consolidated across most measures and, therefore, begin

to provide comprehensive data on all tests routinely employed
in hearing aid evaluation.

The hearing aids employed in this study represented a
limited sample of the range of electroacoustic characteristics.

It would be important to use a larger number of hearing aids
in order to include a greater number of possible electro-

acoustic interactions. A larger sample would also increase

the confidence that regression predictions were made upon a

representative sample of the entire hearing aid population.














CHAPTER 4
SUM~rMARY AND CONCLUSIONS



Audiologists are often called upon to recommend hearing

aids for their clients. The evaluative Drocedures used to

makre these recommendations involve the client's performance

on behavioral measures designed to estimate and quantify

speech intelligibility. It is documented in the research

literature that electroacoustic characteristics of hearing

aids canl affect speech intelligibility, however, there is

disagreement as to which characteristics have th~e mrost profound
affect. Reasons conjectured for the disagreement include

(1) behavioral measure unreliability, (2) inadequate

assessment of electroacoustic interactions, and (3) use of

behavioral measures which are affected by different electro-

acoustic characteristics and/or interactions.

In order to test the validity of these conjectures

forty-two discrete electroacoustic measurements were made on

each of ten hearing aids. Ten behavioral measures C .I.D.

W-22 Wiord Lists (CID), V.A. Discrimination Test (VAH~),

Modified Rhyme Test (MR~T), a synthetic sentence test (SST),

C.I.D. Everyday Sentences: judged intelligibility (ESJ) and

write down~ response (ESW), a ouality judgment passage (QJ) and

the Intermodulation Distortion Test at three S/N~s (IMfDTO, -4,-8))







were recorded through the same aids, degraded with speech

spectrum noise and presented to ten no-rmal hearing listeners.

Except for the synthetic sentence test, C.I.D. Everyday
Sentences: write down response and the Intermodulation

Distortion Test (-83 dB) all of the behavioral measures had

coefficients of reliability > *75 and were, therefore,

concluded to be stable estimates of speech intelligibility.

High inter-correlations between electroacoustic

measurements was interpreted to imply that characteristics

are not truly independent and each represents the moment-to-

moment interactions with all of the other characteristics.

The supposition, therefore, that interaction effects upon

speech intelligibility reouires a systematic appraisal was

confirmed. Assessment of unitary characteristics, apart from

interactions with other characteristics, was determined to be

unrealistic and, perhaps, misleading.

Five factors labelled harmonic distortion, maximum power

output, bandwidth, gain and regularity of frequency response
were extracted from the electroacoustic measurements.

Comparison of beta weights produced by a multiple regression
between the five factors and each behavioral measure showed

that each behavioral measure wras affected by the factors in a

different order and with different strength. The supposition

that behavioral measures are affected by relatively different

electroacoustic interactions was, therefore, confirmed in

these data. Comparison of research results, 'oased upon

different behavioral measures, was decided to be inappropriate

and, maybe, misleading, because the sensitivity of different







measures to electroacoustic interactions do not appear to be

equivalent.

On the basis of the multiple regression analysis, cross

validation and reliability coefficients, the Veterans

Administration Discrimination Test, the Modified Rhyme Test,

C.I.D. Everyday Sentences: judged intelligibility, quality

judgments and the Intermodulation Distortion Test (0 dB) were
determined to be stable and reliable estimates of speech

intelligibility.

An approach to hearing aid selection procedures was

suggested which employed a battery of behavioral measures

made up of the reliable and stable measures in these data.

The approach used knowledge of differential sensitivity of

behavioral measures to electroacoustic interactions to predict

speech intelligibility through a hearing aid.

Finally, implications to further research were outlined

and included replication, cross validation, use of other

behavioral measures, utilization of other hearing aids,

empDloyment of other degradations of the behavioral measures,

and use of subjects having various types and degree of hearing

impairment.




























+cPPENDIX A


FRlEQUENCY RESPONSE CURVES

















FREQUENCY RESPONSE CURVE OF AID NUiMBER TWO
USING THE 2 CC COUPLER





140_ _--- 2a


RESPONSE ---.:-



dB


-L~









1~~T./IREQUENC IN Hzi ~T-



















FREQUENCY RESPONSE CURVE OF AID NUMBER THREE
USING THE 2 CC COUPLER






134_ -- -





RES PONrSE ---
IN -
dB =Li-it-LF








-- r-


84I -~


FREQUENCY IN Hz

















FREQUENCY RESPONSE CURVE OF AID NUMBER POUR
USING THIE 2 CC COUPLER






120_




RES PO N SE i- -~ i - ~ 1 -
IN =_-L~ -?-
dB








-r--?



70_ -L


FREQUENCY IN Hz

















FREQUENCY RlESPONSE CURVE OF AID NUMBER FIVE
USING THE 2 CC COUPLER






1o- E c




RESPONSE
IN ~ ~ ~ I
dB






70~I- --- C-


FREQUENCY IN Hs


















FREQUENCY RESPONSE CURVE OF AID NUMlBER SIX
USING THE 2 CC COUPL~ER





127.5_ p






RES ONSE


dB--- --






50 -F'-


FREQUENCY IN Hz




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs