A?! Eixi‘v‘EES (3‘21“: "EN (3? ti‘x WELHC 4331.??? ‘3‘? 5313229

ASA . - «*JJS'QC’!‘ ”37:3 05' BA ‘LA‘W‘V‘W AND ‘64 TENSITY

 

by. \ Q a. q

2:22:12 '0: 2 1:.» 2.6:} 2:22 of M 4
tr f ,r4! 9.: 6'§‘§.'“F 'g', {(9
113.3% “3.3.92 3’91 1, w.” "W god};

fissxs

LIBRARY

Michigan State
Univcrsity

 

 

 

 

 

.'

 

ABSTRACT

AN INVESTIGATION OF INTELLIGIBILITY
OF SPEECH AS A FUNCTION OF
BANDWIDTH AND INTENSITY

by Lowell J. Sahlstrom

This study investigated the effects of band-pass
filtering and intensity variation upon the intelligibility
of speech. Two bandwidths were employed: the first was
480 ops wide from 1320 cps to 1800 Cps; the second band-
width was 960 CpS wide from 1080 cps to 2040 ops. Both
bandwidths were centered at 1560 cps.

Twenty—four normal hearing adults served as subjects
in this study. The subjects were divided into two groups
of twelve, each of which listened to the test material
under one of the bandwidth conditions. Both groups of
subjects listened to the stimulus at average intensity
levels of 35 dB, 50 dB, and 70 dB sound pressure level.
Each group listened to a practice list at 90 dB SPL before
hearing the experimental stimuli.

The test material utilized in this investigation
were lists of one-syllable words. These lists were re-
corded on magnetic tape from the commercial discs of the
CID W-22 word lists (lists 1E, 2E, 3E, and 4E). Listener's

responses were recorded on prepared answer sheets.

LOWELL J. SAHLSTROM

A two way analysis of variance was performed to test
for significant differences between the variables (intensity
and width of bandpass) involved in this study. The results
of the analysis indicated that there was no significant
difference in intelligibility scores between a 480 ops
bandwidth condition and a 960 ops bandwidth condition, both
of which were centered at 1560 ops. The data obtained
showed that there was a significant difference in intelli-
gibility scores due to hcreaSe in intensity level. It was
also found that there was no significant interaction between
bandwidth and intensity level.

On the basis of this study, the following conclusions
were made:

1. Doubling the size of a 480 ops bandwidth centered
around 1560 ops, does not significantly improve intelligi-
bility.

2. An increase in average intensity level was found
to produce an increase in percent correct recognition of one
syllable words.

3. Bandpass filtering does impair speech intelligi-
bility so that correct responses do not reach one-hundred
percent correct, even at high intensity levels.

Some questions arising from this study were posed as
possible areas for further research relative to speech

intelligibility.

AN INVESTIGATION OF INTELLIGIBILITY
OF SPEECH AS A FUNCTION OF

BANDWIDTH AND INTENSITY

BY

Lowell J. Sahlstrom

A THESIS

Submitted to
Michigan State University
in partial-fulfillment of the requirements
for the degree of

MASTER OF ARTS

Department of Speech

1964

 

TABLE OF CONTENTS

Page
LIST OF TABLES O O O O O O O O O O O O O O O O O O 0 iii
LIST OF FIGURES O O O O O O O O O O O O O O O O O 0 iv
LIST OF APPENDICES . . . . . . . . . . . . . . . . . V
Chapter
I. STATEMENT OF THE PROBLEM . . . . . . . . .' 1
Introduction
Statement of Problem and Purpose
of Study
Hypotheses

Importance of Study
Definition of Terms
Organization of Thesis

II. REVIEW OF THE LITERATURE . . . . . . . . . 10
Summary

III. SUBJECTS, EQUIPMENT, AND PROCEDURES . . . . 30
Subjects
Equipment
Materials
Procedures

IV. RESULTS AND DISCUSSION. . . . . . . . . . . 37

Results and Analysis
Discussion

V. SUMMARY AND CONCLUSIONS . . . . . . . . . . 49
Summary
Conclusions
Recommendations for Further Research

BIBLIOGRAPHY O O O O O O O O O O I V O O O O O O O O O 54

APPENDICES I O O O O I O O C O O O O O O O O O O O O O 57

Table

1.

2.

LIST OF TABLES

Intelligibility Scores (Per cent Correct)
for Filtered Speech . . . . . . . . . . .

Analysis of Type I Design . . . . . . . .

iii

Page

LIST OF FIGURES

Figure Page
1. Schematic Diagram of Stimulus
Presentation Apparatus . . . . . . . . . . . . 33
2. Mean Per cent Scores Under Conditions
47

Of Filtering 0 O O O I O O O O O O O O O O O 0

iv

Appendix
A.
B.

C.

LIST OF APPENDICES

SUBJECTS ANSWER SHEET

INSTRUCTION SHEET

Page
58
60
61

CHAPTER I
INTRODUCTION

Intelligibility of speech has been the subject of
research for many years. -The—first major use of intelli-
gibility tests followed the-development of such tests by
the Bell Telephone Laboratories1 (Fletcher and Steinberg)
during the first quarter of this century. Articulation
or intelligibility curves are now widely employed as
measurements of communication efficiency.

In general the dependent variable has been the per-
centage of correct recognition. The independent variable
may have taken the form of overall gain of the communi-
cation channel. Other parameters which have been involved
in such tests include frequency band width and location,
signal to noise ratio, communication channel distortion,
dialectical differences between speakers and listeners,
hearing acuity, and type of speech material utilized in

the presentation.2

 

1Lee E. Travis, et al. Handbook of Speech Pathology
(New York: Appleton Century CrofEs, Inc.,'19577, p. l4lii

2H. FletCher, and J. E. Steinberg, "Articulation
Testing Methods," Bell Systems Technical Journal_ 8 (1929),
850-854. "

There are several aspects to perception of speech
and one of the important ones is the process that allows
a listener to correctly recognize and to record the speech
sounds which are spoken to him. This is the recognition
aspect of speech perception. One method of measuring this
aspect of perception is to have a speaker read a certain
number of words or sounds to a listener who writes the
word or sounds that he thinks he hears.3

The research that has been done in the area of
intelligibility has examined this phenomenon under many
different conditons and situations. In a study done by
Peterson and Subrahmanyan,4 intelligibility was investi-
gated under conditions of a narrow band speech transmission
system which scanned the time-frequency plane in a sinu-
soidal manner across a range of 6800 ops.

5 investi-

Another study done by French and Steinberg
gated intelligibility using low pass and high pass filters
which filtered out portions of the frequency spectrum

above-and/or below a given point on the frequency scale.

 

 

3Harvey Fletcher, Speech and Hearin in Communication
(New York: D. Van Nostrand Co., Inc., I953), p. 278.

4G. Peterson and D. Subrahmanyan, "Evaluation of Time-
Frequency Scanning for Narrowband Speech Transmission,"
Journal of the Acoustic Society of America, 31 (1959), 113.

5N. R. French and J. Steinberg, "Factors Governing
Intelligibility of Speech Sounds," Journal of the Acoustic
Society of America, 19 (1949), 90.

 

 

Kryter6 reported a study on methods-for arriving at
an Articulation Index for use in estimating the intelli-
gibility potential of electronic instrumentation without
having to use elaborate intelligibility tests. Other re-
search performed by Huizing, Kruisinga, and Taseloar7
examined speech reception under conditions ofgroupings
of band pass filters, giving a wide range of frequency
response.

In a study done by Egan and Weiner,8

groups of band
pass filters were used with two types of masking noise
and a test presentation of nonsense syllables to evaluate
intelligibility.
A study reported by Hirsh, Reynolds, and Joseph,9
investigated the intelligibility of different speech
materials.

This research then, has investigated the intelli-

gibility of speech under many environmental listening

conditions such as noise, limited frequency, various kinds

 

6Karl Kryter, "Methods for the Calculation and Use
of the Articulation Index,“ Journal of the Acoustic Society
of America, 34 (1962), 1689.

7H. S. Huizing, R. J. Kruisinga, and M. Taseloar,
"Triplett Audiometry: An Analysis of Band Discrimination
in Speech Reception," Acta Oto-laryngology, 51 (1960), 256.

8J. P. Egan and G. Weiner, "On the Intelligibility
of Bands of Speech in Noise," Journal of the Acoustic So-
ciety of America, 18 (1946), 435.

 

 

91. J. Hirsh, E. G. Reynolds, and M. Joseph, "Intel—
ligibility of Different Speech Materials," Journal of the
Acoustic Society of America, 26 (1954), 530.

 

of distortion of sound, and different speech materials.
These studies have used high pass filters, low pass filters,
and band pass filters in groupings.

In different ways with different materials and
equipment, intelligibility has been examined quite tho-
roughly. However, none of this research was oriented
towards examining intelligibility with a single band pass
filter in quiet conditions. Furthermore, this does not
appear to have been done with monosyllabic phonetically
balanced words, the type of speech material generally
used in audiometric testing today.

The research that has been done relative to intel-
ligibility has often contributed much toward the goals
discussed in this introduction. In the present study,~
one more aspect of this problem has been investigated in
the hOpe that still more information might be gained in

this area.

Statement~of Problem and Purpose of Study

The purpose of this study was to investigate the
relationship between intelligibility and: (1) width of
bandpass, and (2) intensity. This investigation utilized
two conditions of filtering with a band-pass filter: a
bandwidth of 480 ops; and a bandwidth of 960 cps. Both

bands were centered at 1560 ops.

This study-analyzed and compared the-results obtained
fromrnormal hearing subjects as they responded to the CID
Auditory Test W-22 (lists 1E, 2E, 3E, and 4E.) under each
condition of filtering and at three different levels of
intensity within each bandpass condition.

In order to analyze the difference in intelligibility
found between the two bandwidths, the discrimination scores
of the subjects responding to the test material under the
condition of narrow band pass filtering were compared with
the discrimination score of the group of subjects who re-
sponded to the test material under the wider band pass
condition. Secondly, the diScrimination scores for each
intensity were compared with that of the other two inten-
sity levels for each bandwidth in order to determine dif-
ferences in intelligibility due to changes in stimulus
intensity. Thirdly, the scores were analyzed for any

interaction between width of bandpass and level of intensity.

Hypotheses

 

The following null hypotheses were tested to investi-
gate the differences in intelligibility due to: .(1) width
of bandpass; (2) variations in intensity, and; (3) inter-
action of these factors:

1. There is no significant difference in intelli—

gibility scores derived from listening to a band

of frequencies 480-cps wide and a band of

frequencies 960 ops wide with both bands centered
at 1560 ops.

2. There is no significant difference in intelli-
.gibility scores obtained with filtered speech
due to increase in intensity level.

3. There is no significant interaction effect between

width of band-pass and intensity levels.

Importance of Study

An extensive amount of research has been done in the
area of intelligibility. Much of this research has been
directed toward a determination of the effects of noise on
intelligibility. Other studies have attempted to determine
how much of the speech range of frequencies can be filtered
out of the presentation of various types of speech materials
and still maintain a sufficient degree of intelligibility
for normal communication. Research of this type has impli-
cations for hard-of-hearing persons, for communication
under conditions of excessive noise, and for communication
under limited frequency conditions due to other causes.

The present study prOposed to investigate intelli-
gibility within a narrow frequency range of speech, namely
that range from 1080 ops to 2040 cps. It was believed
that the results of this study would contribute to the
available information regarding the phenomenon of intelli-

_gibi1ity under limited frequency conditions, without the

presence of any masking or other type of noise. Secondly,
these results would yield further information regarding
the influence of intensity variations upon intelligibility

of one syllable words.

Definition of Terms

 

Decibel.--A relative unit of measurement of sound
intensity expressed in a logarithmic ratio of intensity

differences.

Normal Hearing Adults.--Fourteen graduate and ten
undergraduate students, all demonstrating pure tone air
conduction thresholds at 10 dB (re: audiometric zero)
or better for the frequencies 500, 1000, and 2000 ops
in both ears, when tested in a sound treated room with

.a portable pure tone audiometer.

Intelligibility.--The accuracy with which speech is
received by the listener, after having been passed through

a communication channel.

Speech Discrimination or Articulation Test.--A test
which allows for evaluation of an individual's ability to
discriminate between acoustically similar words or words
that contain acoustically similar sounds. It is used to
determine a person's discrimination loss for speech, and
is usually designated as a percentage of the total list

which the person records correctly.

Phonetically Balance§.--Test items in which all,
or nearly all, of the language phonemes are represented.
The frequency of occurence of these fundamental sounds

is in prOportion to their distribution in normal speech.lO

CID Auditory Test W-22.--A discrimination test con-

 

sisting of four lists of 50 phonetically balanced, one-
syllable words adapted by the Central Institute for the
Deaf11 from the Phonetically Balanced Word Lists (PB-50)
of the Psycho-Acoustic Laboratory, Harvard University.12

Each list has been scrambled to make six different word

orders.

Bandspass Filter.--A filter is an electrical circuit

 

which transmits with only a small loss, certain frequencies
or bands of frequencies, and provides attenuation for

other frequencies. A band-pass filter transmits frequencies
in the interval between a low and a high cutoff frequency.13
The band-pass filter used in the present study is a one—

tenth octave filter with a rejection rate of 30 dB per

octave .

 

10James P. Egan, "Articulation Testing Methods,"
Laryngoscope, 58 (1948), 957.

11Ira J. Hirsh, et al., "Deve10pment of Materials
for Speech Audiometry," Journal of Speech and Hearing

Disorders, 17 (1952), 328-329.
12

 

Egan, op. cit., p. 961.

13 . .
TraV1s, 0p. c1t., p. 133.

Bandwidth.--A frequency band expressing in cycles

 

per second, the range of frequencies between the low and
high cutoff points of a band-pass filter. The present
study used two bandwidths; 1320-1800 cps, and 1080-2040

cps; both of which had a center frequency of 1560 cps.

Intensity.--The pressure level of the test material

 

presentation measured in decibels relative to .0002 dynes

per square centimeter.

Organization of the Thesis

 

Chapter I has included an introduction to the
problem of intelligibility and an outline of the purpose
of this study. It has set forth the hypotheses to be
tested, the importance of the study, a definition of
terms used, and an outline of the thesis.

Chapter II presents a review of the literature
pertinent to the present study of intelligibility.

Chapter III describes the subjects, equipment,
materials, and procedures utilized in this study.

Chapter IV consists of an analysis and discussion
of the results of the study.

Chapter V contains the summary and conclusions

of the study, and implications for further research.

CHAPTER I I

Review of the Literature

 

A great deal of research has been done in the area
of intelligibility. .Much of this research has been done
with phonetically balanced words in defining what has
been termed the articulation function. This work has
been done with normal ears as well as with ears having
different types of impairments. Articulation curves
have been constructed, showing how the scores obtained
on articulation tests by normal hearing people and by
peOple with impaired hearing, are affected by the intensity
at which the word lists are presented. Several different
studies have been directed at a determination of the
effects of noise on intelligibility. Other research has
attempted to determine how much of the speech range can
be filtered out of the presentation and still maintain a
sufficient degree of intelligibility for adequate com-
munication.

It remained for telephone engineers interested in
the adequacy of their equipment to develOp procedures
for the quantitative investigation of speech perception.
The concern was with intelligibility rather than percep-

tion, in terms of what the equipment can do rather than

10

11

what the listeners can do. The results were used to
evaluate equipment rather than listeners or speakers.14
Since perception is a psychological aspect of speech,
the telephone personnel were primarily interested in
intelligibility as it related to the performance of
their equipment.

There are several different aspects of speech
perception but the foremost is the process of discrimina-
tion which enables one to recognize correctly and to
record the speech sounds which are spoken. This can be
called the recognition aspect of speech perception.15
The method of measuring this aspect of perception is to
have a speaker read aloud a certain number of speech
sounds to a listener who writes what he thinks he hears.
A quantitative measure of the intelligibility of speech
may be obtained by counting the number of discrete speech
units correctly recorded by the listener in an articula-
tion test. Lists of syllables, words, or sentences are
read aloud and the percentage of items correctly recorded
is called the articulation score.16

The procedure for administering the articulation

tests is to have the listeners record their responses on

 

14S. S. Stevens, (ed), Handbook of Experimental

Psychology (New York: John Wiley and Sons, Inc.,l951hi
p. 1040. '

15Fletcher, op. cit., p. 278.

16James P. Egan, Op. cit., p. 955.

12

a paper with numbered blank spaces. In this way, there
can be little doubt as to whether or not the subject or
listener, has heard the words correctly, and a permanent
record of his responses is available for analysis of his

discrimination errors.17

.This procedure leaves some chance
for inaccuracy due to unclear or poor writing on the part
of the listeners, which produces responses that may be
unreadable.

Articulation tests have proved useful for comparing
communication equipment, for evaluating the effects of
noise on communication, for determining the basic audi—
bility of different words, for hearing examinations, and
for rating and training communication personnel. The
experimental variables are the quality and intensity of
the announcer's or speaker's voice, his pronounciation and
enunciation of words, his accent and proper or imprOper
use of the equipment. Also involved is the phonetic
composition and difficulty of items in the test material.
The communication channel is a factor to be considered
in terms of the noise present in the system, the intensity
of the signal presented, and the overall fidelity of the
equipment. Finally, the listener is another variable in

terms of his hearing acuity, his ability to discriminate

 

l7Hayes Newby, Audiology (New York: Appleton Century
Crofts, 1964), p. 116.

 

l3

speech sounds and his ability to ignore any masking noise.
Thus this three part system of speaker, channel, and
listener contains an extensive list of variables which
are involved in the use of articulation tests in evalu-
ating any aspect of intelligibility. And yet these must
be considered because of the effects each may have on
the results of an articulation test.18

In some of the work done on the relation of inten—
sity to intelligibility, it has been found that the indi-
vidual's threshold for speech (the speech level necessary
in order to identify half the test items correctly) depends
upon the type of speech material in the test. Kryter
reported that a level of 31 dB sound pressure level was
necessary for fifty per cent recognition of monosyllables

19 Davis reported a level of 33 dB SPL

20

in a free-field.

for monsyllables with earphones. In the same study,

Davis found spondees to be reported correctly half the

21

time at a level of 22 dB SPL with earphones. Shaw,

Newman, and Hirsh reported spondees at a threshold level

 

18Egan, 0p. cit., p. 1043.

19K. D. Kryter, "Effects of Ear Protective Devices
on the Intelligibility of Speech in Noise," Journal of
the Acoustic Society of America, 18 (1946), 416.

20Hallowell Davis (ed.), Hearing and Deafness (New
York: Murray Hill, 1947), p. 150.

ZlIbid.

14

of 17 dB SPL monaurally and 14 dB SPL when tested bi-

naurally.22

French and Steinberg found nonsense sylla-
bles to be recorded correctly fifty per cent of the time
at approximately 30 dB.23 Hawkins and Stevens examined
thresholds for connected discourse and found a level of
24 dB SPL when presented by earphones.24

The naive listener who takes an articulation test
for.the first time, will yield initial thresholds which
are several decibels poorer than his later thresholds
will be, particularly if time is not taken to familiarize
him with the test words prior to the initial test. On
the other hand, any person having appreciable prior ex-
perience with the speech material to be used in the test
will obtain thresholds which appear better by several
decibels than they would, had they not had that experience

25

with the material. This would seem to indicate that in

any investigation of intelligibility, it would be advisable

 

22w. A. Shaw, E. B. Newman, and I. J. Hirsh, "The
Difference between Monaural and Binaural Thresholds,"
Journal of Experimental Psychology, 37 (1947), 240.

23

 

French and Steinberg, 0p. cit., p. 117.

24J. E. Hawkins, Jr. and S. S. Stevens, "The Masking
of Pure Tones and of Speech by White Noise," Journal of
the Acoustic Society of America, 22 (1950), 12.

25J. Jerger, R. Carhart, T. Tillman, and J. Peterson,
"Some Relations Between Normal Hearing for Pure Tones and
Speech," Journal of Speech and HearinggResearch, 2 (1959),
139.

 

 

15

to provide a practice session in which the subjects are
allowed to become familiar with the required tasks. In
this way the effects of test familiarity are controlled.
Lists of phonetically balanced words are often
used for articulation tests. Two of the more commonly
known such tests are the Phonetically Balanced Word
Lists of the Psycho-Acoustic Laboratory, Harvard Univer-

26 and the Central Institute for

sity developed by Egan,
the Deaf W-22 lists which were develOped from the PAL-50
lists by Hirsh.27 Both of these tests are composed of
monosyllabic words.

It has been found that as intensity is increased
above the individual's threshold for speech, scores
achieved on phonetically balanced word tests increase
rapidly until the intensity at which the words are
presented reaches a level of about 40 dB greater than
the individual's threshold for speech. At this point the
curve tends to flatten out and there is a negligible
increase in discrimination score beyond this level.

Hirsh and others report a study designed to investi-
gate whether the different lists of the CID W-22 test

were of equal difficulty. They provided a practice run

in which the subjects first listened to one list at a

 

26Egan, op. cit., pp. 955-991.

27Hirsh, 0p. cit., pp. 321-337.

16

level of 100 decibels. This gave the listeners an indoc-

trination period in which the words were presented at a

sufficiently high level for them to become adapted to the

type of task assigned. This study found that there were

no consistent differences between scores that were ob-

tained from listeners responding to the different lists

of the W—22 test.28h
In another study of the differences in difficulty

of the W—22 word lists, Elpern29 accumulated 1490 monaural

discrimination scores from Veterans Administration Audi—

ology Clinics in six major cities. He collected such

scores on each of the four fifty word lists which make

up the W-22 test. Each of these lists are recorded in

six different word orders designated A, B, C, D, E, and

F. Elpern found that there were differences between

these lists. These differences existed in both average

level of difficulty and in average range of difficulty,

among the four word lists. However, these differences

are found to be so small that they would not interfere

with normal clinical usage. In certain research appli—

cations the differences may be great enough to warrant

more careful attention, according to Elpern. He found

that the only lists which do not differ in any respect

 

28Ira J. Hirsh, et al., 0p. cit., p. 333.

29Barry 8. Elpern, "Differences in Difficulty Among
the CID W-22 Auditory Test," Laryngoscope, 70 (1957), 1564.

 

17

from one another are lists 2 and 3; and lists 3 and 4.
He concludes that these pairs may be employed in re-
search situations without danger of unequal difficulty
of word lists being an uncontrolled variable in the
design. Another alternative would be to use but one
of the lists, and to use different groups of subjects
for each treatment.

In considering the equal difficulty of articulation
test, one must look at all variables. If such lists are
to be used in a study which entails conditions of dis-
tortion of any kind, there may be a change in the rela-
tive difficulty of a given list due to the phonetic ele-
ments which are lost as a result of the filtering pro—
ducing the distortion. Thus another variable is intro-
duced which should be investigated. One study was done

1.30 In this investi-

on this subject by Hardick and Dea
gation, speech discrimination under conditions of filtering
was studied. Listeners were asked to respond to the W-22
word lists which were presented to them. The speech sig-
nal was distorted by the use of a low-pass filter. The
results of this study indicated that the four lists of

the W-22 test were of equal difficulty under the adverse

listening conditons that resulted from frequency distor-

tion of the speech signal.

 

3OEdward Hardick‘and Leo Deal, "The Effects of Fre-
quency Distortion on Speech Discrimination" (unpublished
study, Speech Department, Michigan State University, 1962).

l8

Continuous.discourse is representative of speech
encountered in everyday life. There are those who feel
that this would be the best means of testing intelligi-
bility, especially in a clinical situation. Scoring of
such a test presents a much greater problem than word

31 compared intelligi-

lists, however. Giolas and Epstein
bility scores obtained on the W-22 word lists, the PB-50
lists and continuous discourse. It was found that the
monosyllabic words of the W-22 lists yielded higher
scores than the PB-50 lists. Frequency distortion was
produced by the use of low-pass filters. Six different
filtering conditions were presented with the speech
sample being subjected to all six filtering conditions
as well as to full-range reproduction. Listeners re-
sponded to the stimulus presentation by writing their
answers on numbered sheets of paper. The results of this
study indicate that frequency distortion has an effect
on the W-22 lists which is much the same as the effect
on continuous discourse. With both types of speech ma-
terial, errors increased as distortion increased.

In a study of the effect of word familiarity on

32

intelligibility, Owens presented several lists of

 

31Thomas G. Giolas and Aubrey Epstein, "Comparative
Intelligibility of Word Lists and Continuous Discourse,"
Journal of Speech and Hearing Research, 6 (1963), 354.

32Elmer Owens, "Intelligibility of Words Varying in
Familiarity," Journal of Speech and HearinggResearch, 4
(1961), 124.

19

monosyllable words to groups of listeners. Each list was
presented under conditions of frequency distortion caused
by the use of filters. It was found that those words
which are more familiar are significantly more intelli-
.gible. In relating these findings to the PB-50 and W-22
word lists, it was learned that the W-22 words then are
apparently more familiar, easier to understand, resulting
in a higher intelligibility score. This is possibly the
reason that the W-22 lists score closer to continuous
discourse than other word lists.

Black did a study on the relative intelligibility
of words, in which he investigated the relationships
between aspects of the syllabic pattern, word familiarity,

33 These fac-

and the phonetic characteristics of words.
tors were examined both in a quiet environment and under
conditions of noise. It was found that more familiar
words are more accurately identified by listeners in
either quiet or noise conditions. Words containing two
syllables are more intelligible than those of one syllable.
Those words that have the accent on the second syllable

are more intelligible than words having an accent on the
first syllable. Finally, it was found that the more

sounds contained in a word, the more intelligible the

word becomes.

 

33John W. Black, "Accompaniments of Word Familiarity,"
Journal of Speech and Hearing Disorders, 17 (1952), 416.

20

Licklider and Miller34

investigated the possible
effect that interruption of the speech signal would have
upon intelligibility. They varied the rate of the inter-
ruption and the regularity of the interruption. The high
rate of interruption of the signal did not significantly
impair intelligibility. When using a low rate of inter-
ruption of the signal, there was impairment or reduction
of intelligibility. When the interruption is sufficiently
slow that it may remove entire syllables from the pre-
sentation, it is quite likely that articulation scores
will fall as a result. Intelligibility was found to be
approximately prOportional to the amount of time the
signal was present when using the slow rate of interrup-
tion. In varying the regularity of the interruption, it
was found that such interruptions produced little change
in intelligibility, whether the interruptions were regular
or irregular.

Siegenthaler and Hardick35 produced articulation
curves for normal hearing adults as obtained with several
common word lists. Various phonetically balanced word

lists were used as the test material. Children were used

 

34J. C. R. Licklider and G. A. Miller, "The Intelli-
gibility of Interrupted Speech," Journal of the Acoustic
Society of America, 20 (1948), 593.

 

35Bruce Sigenthaler and Edward Hardick, "Intelli-
gibility Scores Using Various Phonetically Balanced Word
Lists," Pennsylvania Speech Annual, (1959), 8.

 

21

in the study in a comparison of the same lists with two
phonetically balanced word lists designed for use with
children. It was found that there was little difference
in difficulty between the W-22 and the PB-50 word lists,
which is of special interest to the present study. The
study indicates that for normal-hearing children and
adults, differences in articulation curves are relatively
small when listening to these phonetically balanced word
lists in a free-field situation.

Hirsh, Reynolds and Joseph36 studied the relative
intelligibility of different speech materials. The ob-
tained articulation scores for nonsense syllables, mono-
syllables, disyllables and polysyllables. The speech
signal was passed through various filtering conditions
before being presented to the listeners. The articula-
tion scores were also studied as a function of the fil-
tering condition using low and high pass filters. The
results of this study indicated that eliminating all
frequencies above 1600 cycles per second did not impair
the intelligibility of speech seriously. Elimination of
all frequencies below 1600 cps also produced very little
reduction in intelligibility. Examination of the data

pertinent to the number of syllables in a word indicated

 

361. J. Hirsh, E. G. Reynolds, and M. Joseph, "Intel-
ligibility of Different Speech Materials," Journal of the
Acoustic Society of America, 26 (July, 1954), 537.

 

22

that intelligibility is a direct function of the number
of syllables in the word. Monosyllables are more intel—
ligible than nonsense syllables; disyllables are more
easily understood than monosyllables; and polysyllables
are more intelligible than disyllables and so on up to
continuous discourse.

Pollack37 investigated the effects of filtering
on intelligibility by having two speakers read lists
of monosyllables under various conditions of filtering,
employing both high pass and low pass filters. Inten-
sity was also varied. All stimuli were presented in a
background of white noise. At low intensities, he found
that the low frequencies are more important to intelli—
gibility than they are at high intensities. That is,
the low frequencies are important when the speech signal
is weak, but not so important with a strong signal.:
During one part of the study, two speakers read lists of
monosyllable words (PAL and PB-SO word lists) at a level
of 68 decibels re: .0002 dynes per square centimeter.
It was found that frequencies above 2375 cps contributed
little to the overall intelligibility of speech at this
intensity level. Under the same condition of intensity,

frequencies below 425 cps also had little effect on

 

37Irwin Pollock, "Effects of High Pass and Low Pass
Filtering on the Intelligibility of Speech in Noise,"
Journal of the Acoustic Society of America, 20 (1948), 265.

 

23

intelligibility. In general, it was found that intel-
ligibility increased as the bandwidth and the intensity
level of the speech signal were increased.

38 were interested in the

Harris, Harris, and Myers
importance of the 3000 ops area of the frequency spec-
trum for the understanding of speech. They used speech
which was unfiltered, but distorted the speech signal
through a speed-up in words-per-minute. This signal was
presented to persons whose audiograms were similar to
that which would be obtained with low pass filtering at
a cutoff frequency of 3000 ops. Sentence intelligibility
tests were used as the test material. It was found that
a normal or near normal pure tone audiogram in the region
of 3000 cps is essential for high sentence intelligi-
bility when the signal is distorted through a speed-up
in words-per-minute. These results indicated that about
fifteen per cent of the cues for sentence intelligibility
are dependent upon the frequencies around the 3000 cps
region. With a slower speed of the speech signal, at or
near normal speeds in terms of words per minute, these
lost cues can be compensated for by the redundancy of
cues in normal speech. When the speech is speeded up as

in this study, the listener is unable to pick up enough

of these clues to maintain intelligibility.

 

38J. D. Harris, L. Harris, and C. K. Myers, "The
Importance of Hearing at 3KC for Understanding Speeded
Speech," Laryngoscope, 70 (1957), 144.

 

24

French and Steinberg39 report an extensive study
of the factors that influence the intelligibility of
speech. They used filtering of various kinds upon a
speech signal emitted by both male and female voices.
The results indicated that articulation scores increased
as gain was increased. Here again, it was found that
the low frequencies contribute little to intelligibility,
despite the fact that they carry most of the speech power.
When all frequencies above 1900 cps were eliminated; and
when all frequencies below 1900 cps were eliminated,
approximately the same articulation scores were achieved.
The authors described a measure called the "Articulation
Index" which can be computed from the intensities of
speech received by the ear as a function of frequency.40
This provides an easier method for determining the effi-
ciency of a communication system. That is, one can tell
from a stated value of the Articulation Index, the degree
of intelligibility one can receive from a given system
without the necessity of completing articulation tests

with the apparatus using trained subjects.

 

39N. R. French and J. Steinberg, "Factors Governing
Intelligibility of Speech Sounds," Journal of the Acoustic
Society of America, 19 (1949), 117.

4OIbid., p. 109.

25

A study reported by Licklider41

investigated the
effects of amplitude distortion upon intelligibility.
The primary type of amplitude distortion used was peak
clipping. In this process the points of greatest ampli-
tude are clipped off so that the speech signal loses
some of its phonetic elements. Doing this in quiet
situations it was found that peak clipping, even with
the speech signal reduced to one-tenth of its original
amplitude, did not cause much decrease in intelligibility.
Listeners reported ninety-six per cent of the words cor-
rectly under these conditions. The same procedure car-
ried out under conditions of noise produced almost com-
plete loss of intelligibility. Two different types of
noise conditions were used by changing the.looation of
the peak clipping apparatus within the system. In the
first, the clipper was inserted so that both the noise
and the speech were clipped. In the second, the clipper
was inserted in the system so that only the speech sig-
nal was clipped and the noise retained its original am-
plitude.

The intelligibility of bands of speech in noise

was studied by Egan and Weiner.42 They utilized four

 

41J. c. R. Licklider, "Effects of Amplitude Dis-
tortion upon the Intelligibility of Speech," Journal of
the Acoustic Society of America, 18 (1946), 430.

42J. P. Egan and F. M. Weiner, "On the Intelligi-
bility of Bands of Speech in Noise," Journal of the Acous-
tic Society of America, 18 (1946), 439.

 

26

'groups of filters with varying bandwidths within the
group of filters but with each filter in a given group
having a common center frequency. Each of the four
groups had a different center frequency and band width.

A masking noise was presented with the speech signal

at all times. Two types of masking noise were used;

one in which the noise was filtered along with the speech
signal; and a second in which the noise was not filtered.
Nonsense syllables were used as the test material and
scored as per cent correct. The results of this investi—
gation were plotted in terms of equal-articulation curves,
yielding a family of such curves. This provided a system
by which one can determine from this graph, the amount

of increase in intensity needed to counteract or to com-
pensate for a given decrease in the bandwidth of a speech
signal presentation. Of special interest here is that it
was found that for constant levels of received speech,
the articulation score obtained with wide bands of speech
is always higher than the articulation score obtained
with narrower bands of speech.43 This would seem to in-
dicate that the wider the bandwidth, the better the articu-

lation score which will be obtained.

 

42J. P. Egan and F. M. Weiner, "On the Intelligibil-

ity of Bands of Speech in Noise," Journal of the Acoustic
Sociepy of America, 18 (1946), 439.

43

 

Ibid., p. 440.

27

Kryter used band pass filters in combinations of
one, two, or three filters, and using various center fre-
quencies to study the effects of speech bandwidth com-
pression.44 This investigation also utilized unfiltered
conditions. The study used both phonetically balanced
words and sentence intelligibility tests. It was found
that one bandpass filter with wide band limits will give
intelligibility scores which are nearly double that
achieved with several bandpass filters at narrow settings
of bandwidth but with the bands adjacent to each other.
The results of the study indicated that the region around
1600-1700 cps appeared to contribute the most to speech
intelligibility when speech was filtered with a single
pass band system.45 Kryter states that:

Presumably then it is still an open question as to

how much and what protions of the speech spectrum

gzgoaeailimigateglbeforeliggelligibility is reduced
p a e eve .

In this study, Kryter examined the effects of speech
bandwidth compression under conditions of noise. However,

the question pertinent to the present study relates to

the same factors in quiet conditions.

 

44K. D. Kryter, "Speech Bandwidth Compression Through
Spectrum Selection," Journal of the Acoustical Society of
America, 32 (1960), 547.

4SIbid., p. 549.

461bid., p. 547.

28

Another study accomplished by Kryter led to the
determination of twenty frequency bands of speech which
produced equal contribution to the intelligibility of
speech.47 One purpose of this study was to develop an
improved method of calculating the Articulation Index
developed by French and Steinberg.

Harris and Brown4 8

subjected a speech signal of a
sentence intelligibility test to several types of dis-
tortion while filtering with a variable bandpass filter.
They studied the interaction of distortion and filtering
upon the intelligibility of the speech signal. The types
of distortion employed in this study were; shouting,
interruption of the signal, and reverberation. One type
of distortion, that of reverberation, caused a reduction
in intelligibility of eighty per cent. With the other
types of distortion and with speech filtered so that at
the widest bandwidth used in this study, that of 1500
cps, the intelligibility of key words in a sentence
intelligibility test remained high. This band was

centered around 1600 cps. Despite the introduction of

distortion, an articulatiOn Score of ninety-five per cent

 

47K. D. Kryter, "Methods for the Calculation and Use
of the Articulation Index," Journal of the Acoustical
Society of America, 34 (1962), 1691.

48J. D. Harris and L. W. Brown, "Interactions Among
Bandwidth, Center Frequency, and Type of Distortion in
Speech Intelligibility," Journal of the Acoustic Society
of America, 34 (1962), 1999.

 

 

29

was obtained. At the narrowest bandwidth used, (500 cps)
the optimum center frequency for all distorted speech
shifted upward to 1900 cps or above. With the lowest
center frequency utilized in this study, 800 cps, intel-
ligibility remained just as good as at the next higher
center frequency of 1200 cps when both bands were at a

width of 500 ops.

Summary

This survey of the literature relevant to speech
intelligibility has illustrated the extensive amount of
research that has been performed in this area. As a result
of these studies, much is now known about the effects of
such factors as; distortion, filtering, bandwidth com-
pression, different speech materials, increased speech
signal speed, and noise upon the intelligibility of speech.
This survey also reveals, however, that intelligibility of
monosyllable words has not been examined under a single
bandpass filtering condition in a quiet environment, and
as a function of intensity. The present study is an
attempt to fill this gap in the research on speech intel-
ligibility and to add to the fund of knowledgein this

area.

CHAPTER III

SUBJECTS, EQUIPMENT, MATERIALS, AND PROCEDURES

Subjects

The subjects participating in this study were
twenty-four male and female students majoring in speech
attending Michigan State University. There were four-
teen graduate and ten undergraduate students. Each
subject demonstrated a hearing level of 10 dB re: nor-
mal threshold or better in both ears, at the frequencies
of 500, 1000, and 2000 cycles per second. The hearing
level was determined by testing each subject with a
portable pure tone audiometer (Beltone, Model lO-C) in
a sound treated room. No individual was used as a sub-
ject in this study who had extensive experience with

the CID W-22 word lists.

Equipment

 

Pure-tone portable audiometer (Beltone, Model lO-C)
Tape recorder (Ampex, Model 601)

Phonograph (Thorens)

Filter (Allison, Model 25)

Electronic Mixer (Ampex, Model MX-35)

Binaural earphones (Telephonics, Model TDH-39)
Electronic Voltmeter (Bruel & Kjaer, Type 2409)

30

31

Materials

 

CID Auditory Test W-22 (Lists 1E, 2E, 3E, and 4E)
Magnetic recording tape (Audio Master acetate)
Forms for recording responses by subjects (see

Appendix B)

Procedure

 

Preliminary procedures.--The commercially available

 

disc recordings of the CID Auditory Test W-22 were used.
These recordings were transcribed onto magnetic tape.

The discs were played on a phonograph (Thorens). The
output signal was fed directly to the input circuit of

a tape deck (Ampex, Model 601). The calibration tone

on the recordings was set to zero level on the phonograph
output VU meter. The same calibration tone was tran-'
scribed onto the tape recording at a level of ~10 on the
tape deck VU meter. The phonograph output was adjusted
to a level of 85 dB re: audiometric zero. The four CID
W-22 word lists, 1E, 2E, 3E, and 4E were then transcribed
onto the magnetic tape with a fifteen second pause between
each list.

A voltmeter was used to determine the output levels
of the stimulus material, using the dbm scale of the meter.
Output of the amplifier was converted from the dbm scale
as sound pressure level by the use of the following formula:

SPL = 130 + A - M — 20*; where M = meter range, dbm scale;

32

A - meter reading; and * = a correction factor for trans-
ducer loss previously derived through the use of white
noise.

The calibration tone of the tapes of the CID lists
was set at specified levels prior to the actual presenta-
tion. Using the above formula, output of the amplifier
was adjusted to the four desired levels for each bandwidth
condition and the output controls of the amplifier were
marked at these points. In this way output levels could
be easily changed to the necessary new levels during the
presentation. Such settings were determined for average
levels of 35 dB, 50 dB, 70 dB, and 90 dB, re: .0002
dynes per square centimeter. The electronic apparatus
used in the presentation of this study was set up in the

manner illustrated in Figure l.

Presentation.--The presentation of the test material

 

was carried out in the Speech and Hearing Science Labora-
tory of Michigan State University. The walls and ceiling
are plastered; the floor is tiled. The room is equipped
with twelve desk-type chairs, each of which is provided
with binaural earphones connected to the output of the
amplifier. On week-days the ambient noise present due to
classes and activity in other parts of the building is
relatively high. To eliminate much of this noise, this
study was conducted on a Saturday morning when there was
little activity in other parts of the building and sur-

rounding area.

33

mmconmhmm HMHDmCAm

 

Y Y Y Y

 

Y Y Y \r
‘tH/Yv

 

 

Hmflwflamfim
InmxHE

Decoupomam

 

, [

 

 

 

Emummm

Hansen

 

mmmmlccmm

msumnmmmd GOAHMpcmmmHm msHsEHum mo Emummflo oeumfiocomnl.a .mHm

 

 

 

Hommcmuu
was“

oaumcmmz

 

 

34

The twenty-four subjects were randomly divided into
two groups of twelve persons each. Each group received a
separate treatment or condition of the study. As each
group entered the room, they were seated in the desks pro-
vided for them. Answer sheets were provided for each
subject. Each answer sheet consisted of two sets of fifty
numbered blank spaces. Two answer sheets were given to
each subject (see Appendix B). The subjects were then
given instructions appropriate to the task they were

being asked to perform (see Appendix C).

Testing, group I.——The band-pass filter was ad-

 

justed to pass a band of 480 cps from 1320 to 1800 cps
with a center frequency of 1560 cps. Since the output
levels had been previously calculated, it was necessary
only to set the master output control of the mixer to

the setting desired. The first intensity level for Group
I was 90 dB. The filter setting was accomplished by
setting both the low and high cutoff controls at 1200
cps. The low cutoff multiplier was set at 1.1 and the
high cutoff multiplier at 1.5.

The subjects were told to place the earphones on
their ears and prepare to respond as instructed. The
tape transport was turned on and allowed to run through
the entire first list of words. The operator listened to
the words through a monitor headphone as a check on the

operation of equipment and as a guide to progress of the

35

list. The first was considered a practice list. During
the fifteen second pause between lists on the tape, the
output setting on the master control of the mixer was
re-adjusted to the 70 dB setting. The subjects had time
to turn to the next list of numbered blank spaces. The
tape was allowed to play through the second list. During
the pause between lists the master control was adjusted
to the 50 dB setting. Again, the tape was allowed to
continue through the third list of fifty words. Once
again the output control was adjusted during the pause
between lists, to the 35 dB setting. At the end of this
list, the tape recording was re-wound to the beginning
of the first list and the equipment power supply was
switched off. Subjects responses were collected and the

subjects dismissed.

TestingyrGroup_II.--In testing the second bandwidth

 

condition, again both the low cutoff and high cutoff con-
trols of the filter were set at 1200 cps. The low cutoff
multiplier was set at .9 and the high cutoff multiplier
was set at 1.7. This yielded a band pass of 960 cps from
1080 to 2040 cps, with a center frequency of 1560 cps.
The master output control of the mixer was again set at
the 90 dB setting.

The subjects were seated, given the instructions
and answer blanks, and told to place the earphones and

prepare to respond as instructed. The tape transport was

36

turned on as before. The presentation of material and
output settings were adjusted during pauses between lists
in the same manner as described for Group I. The first
list presented at 90 dB again served as a practice list
only. At the end of the fourth list, the equipment was
switched off and the listeners answer sheets collected

and the subjects dismissed.

Summary.-—Group I listened to four lists of W-22
words under a 480 cps bandpass filtering condition.
Group II listened to the same four lists of W-22 words
under a 960 cps bandpass filtering condition. Both
bandwidths were centered at 1560 cps. Both groups
listened to list 1E at 90 dB, which was used as a prac-
tice list. List 2B was presented to both groups at 70 dB,
list 3E at 50 dB, and list 4E at 35 dB, all intensity

levels with reference to .0002 dynes per square centimeter.

CHAPTER IV

RESULTS AND DISCUSSION

Introduction

 

One of the problems investigated in this study
was whether there was any significant difference in
intelligibility scores obtained from subjects listening
to two narrow but different bandwidths of speech. A
second question sought to determine whether there was
any significant increase in intelligibility due to in-
crease in stimulus intensity. Thirdly, the question was
asked regarding the presence of any significant inter-
action between bandwidth and intensity.

The quantification of responses was done for each
of the two filtering conditions at each of the three
intensity levels. The per cent correct intelligibiltiy

scores for filtered speech are illustrated in Table 1.

Analysis

A two-way analysis of variance was done to deter-
mine whether there was any significant difference between
bandwidths, betWeen intensity levels, and whether there

was any significant interaction between bandwidth and

37

38

TABLE 1

INTELLIGIBILITY SCORES (PER CENT CORRECT)
FOR FILTERED SPEECH

 

Subject 35 dB* 50 dB* 70 dB* 90 dB*

 

Within the 1320 to 1800 cps Bandwidth Condition

 

 

 

l 26 74 86 90
2 18 74 76 88
3 20 60 74 78
4 16 76 72 76
5 18 60 7o 74
6 26 60 72 68
7 16 58 72 60
8 18 68 74 58
9 20 56 66 50
10 18 62 68 48
11 4 56- 60 26
12 2 52 80 , 6
x 16.83 63.0 72.5 60.16**

 

Within the 1080 to 2040 cps Bandwidth Condition

 

 

 

 

l 24 70 82 88
2 16 58 84 78
3 28 72 80 78
4 18 60 82 72
5 18 56 62 68
6 10 66 74 64
7 10 52 66 62
8 14 46 68 60
9 24 44 66 54
10 20 56 74 50
ll 12 44 56 46
12 16 52 78 36
X 17.5 56.3 72.6 63.0**
* Intensity levels re: .0002 dynes per square centimeter.

** Denotes practice lists.

39

intensity level. A two-factor Type I mixed design as

49 was utilized for this analysis.

described by Lindquist

In the statistical analysis summarized in Table 2,
the B effect represents the differences in intelligibility
scores obtained between bandwidths. ‘The A effect repre-
sents the differences in intelligibility scores obtained
between intensity levels. The AB effect represents the
interaction between bandwidth and intensity levels.

The analysis in Table 2 indicates that there is no
significant difference in intelligibility scores at the
five per cent level of confidence between bandwidths used
in this study. Presenting one bandwidth 480 cps wide and
a second bandwidth 960 cps wide results in no significant
difference in intelligibility scores. Therefore, the
first hypothesis that there is no significant difference
in intelligibility scores between a 480 cps bandwidth
condition and a 960 cps bandwidth condition, both of
which have a center frequency of 1560 cps, cannot be
rejected.

The second hypothesis states that there is no sig-
nificant difference in intelligibility scores due to
increase in intensity level. The analysis in Table 2

shows that there is significant difference in intelli-

gibility scores between intensity levels at the five

 

49E. F. Lindquist, Design and Analysis of Experiments
in Psychology and Education (Boston: Houghton Mifflin Co.,
1953), p. 267. .

 

40

TABLE 2

ANALYSIS OF TYPE I DESIGN

 

 

 

 

 

 

Source df Sums of Squares Mean Squares F
Between
Subjects 23 2617.278

B
(Bandwidth) 1 68.055 68.055 .587*
Error (b) 22 2549.223 1 115.873
Within
Subjects 48 42016.00

A
(Intensity) 2 40352.776 20176.388 607.320**

AB
(Interaction) 2 201.446 100.723 3.03.l=
Error (w) 44 1461.778 33.222

4 w

Total 71 44633.278

 

* An F of 4.30 is required for significance at the 5 per
cent level of confidence, with df = 1, 22.

** An F of 3.214 is required for significance at the 5 per
cent level of confidence, with df = 2, 44.

= An F of 3.214 is required for significance at the 5 per
cent level of confidence, with df = 2, 44.

41

per cent level of confidence. Increasing the intensity
level of the speech signal from 35 dB SPL to 50 dB SPL

to 70 dB SPL under both bandwidth conditions results in
an increase in intelligibility scores. Under the 480

ops bandwidth condition this change in intelligibility
scores increased from about 16 per cent correct at 35 dB
to 63 per cent correct at 70 dB. Under the 960 cps band-
width condition there was an increase from 17.5 per cent
correct at 35 dB to 56.3 per cent at 50 dB, to 72.6 per
cent correct at 70 dB. Since there is significant dif-
ference in intelligibility scores as a function of inten-
sity, the second hypothesis can be rejected.

The analysis of the AB effect in Table 2 (inter-
action between bandwidth and intensity level) reveals no
significant interaction between bandwidth and intensity
levels. Therefore, the third hypothesis which states
that there is no significant interaction effect between

width of bandpass and intensity levels, cannot be rejected.

Discussion

 

The analysis of the data obtained in this study
indicated that the first hypothesis regarding differences
in intelligibility scores obtained from two bandwidth con-
dition, could not be rejected. The two bandwidths used
in this study were one of 480 cps and another of 960 cps.
The first included the range from 1320 to 1800 cps. The

second enclosed the range of frequencies from 1080 to

42

to 2040 cps. Both bandwidths had a center frequency of
1560 cps. The first bandwidth is only one-half as wide
as the second bandwidth.

The analysis indicated, however, that there was
no significant difference in intelligibility scores ob—
tained under the two conditions. This would indicate
that under the conditions of this study, intelligibility
of speech was not impaired any more seriously by the
narrowest bandwidth condition than it was by the widest
bandwidth condition. Apparently enough information is
transmitted through a 480 cps bandwidth centered around
1560 cps so that listeners are able to discriminate
speech fully as well as they can when the bandwidth is
twice as wide.

This finding is not in agreement with a study
reported by Egan and Weiner50 and reviewed in Chapter II
of this thesis. They reported that intelligibility
scores obtained from wide bandwidth conditions are always
higher than those obtained from narrower bandwidths. This
does not specify, however, how much wider that bandwidth
must be in order to reach a significant increase in
intelligibility scores.

One possible explanation for this inconsistency

is that both bandwidths that were used in the present

 

50Egan and Weiner, op. cit., p. 439.

43

51 and Harris and

study were centered at 1560 cps. Kryter
Brown52 have reported in separate studies that the region
around 1600 cps is the most important area for speech. It
appears from the present study that enough phonetic clues
are carried in this frequency region so that more than
double the width of a 480 cps band of frequencies is re-
quired to produce a significant increase in intelligi-
bility scores from one bandwidth condition to another.
Filtering such as that used in this study, produces
frequency distortion of the speech signal. Such dis-
tortion is known to cause a decrease in intelligibility
scores. In the report of Giolas and Epstein53 reviewed
in Chapter II, it was concluded that errors in intelligi-
bility tests increase as distortion is increased by fil-
tering. Any increase in distortion caused by filtering
will produce a decrease in intelligibility scores. Such
a decrease in intelligibility scores did not occur in
the present study. One would expect that the 480 cps
bandwidth would have more errors and thus a lower intel-
ligibility score than the 960 cps bandwidth condition,
according to the Giolas and Epstein study. Under the
conditions of the present study, increase in distortion

did not cause an increase in errors of intelligibility.

 

51Kryter, 0p. cit., p. 547.

52Harris and Brown, 0p. cit., p. 1999.

53Giolas and Epstein, loc. cit.

44

The widest bandwidth used in this study was twice as
wide as the narrow bandwidth. There was no significant
difference in intelligibility scores between the two band-
widths. This raises the question of how much wider the
widest bandwidth of frequencies must be in order to pro-
duce a significant difference in intelligibility scores.

At the center frequency of 1560 cps, doubling the
width of a 480 cps frequency band does not produce a
significant difference in intelligibility scores. One
is led to ponder what the results of a similar study
would be if the center frequency of two like bandwidths
were shifted either upward or downward on the frequency
spectrum. For example, would there still be no signi-
ficant difference in intelligibility scores for a band-
width twice as wide as a second bandwidth if the center
frequency for both bands were at 2500 cps, or at 1000
cps? Secondly, would the ratio of one bandwidth to another
in terms of the needed increase in bandwidth to produce
significant difference in intelligibility, be the same
as it is for a center frequency of 1560 cps? Or would we
find that this ratio would change at other points on the
frequency spectrum?

The second hypothesis which dealt with differences
in intelligibility of speech due to changes in intensity
levels was rejected at the five per cent level of confi-
dence. Three intensity levels were used in this study;

35 dB, 50 dB, and 70 dB sound pressure level.

45

Normal hearing adults will usually achieve a fifty
per cent recognition score at about 22 dB sound pressure
level, with monosyllable words in unfiltered conditions.54
Under the filtering conditions used in the present study,
intelligibility scores reached just over fifty per cent
at the fifty decibel intensity level, twenty-eight deci-
bels greater than that obtained under unfiltered conditions.
At the lowest level of 35 dB, 13 dB above expected unfiltered
speech reception threshold for the W-22, intelligibility
scores obtained in this study did not reach twenty per cent
correct.

Under unfiltered conditions, normal hearing adults
can be expected to achieve a one hundred per cent correct
intelligibility score at approximately speech reception
threshold plus forty dB or at about sixty-two dB sound
pressure level. In the present study, however, at the
highest intensity level used (70 dB) intelligibility scores
only reached a level of about seventy-two per cent correct,
or twenty-eight per cent below what would be expected
under unfiltered conditions for normal subjects.

The data in Figure 2 shows a large increase in
intelligibility scores from 35 dB to 50 dB sound pressure
level, but a much smaller increase from 50 dB to 70 dB
sound pressure level. There is an increase in intelli-

gibility scores due to increase in intensity. This is in

 

54Hirsh, et al., 0p. cit., p. 334.

46

agreement with a study reported by French and Steinberg55

and reviewed in Chapter II, in which it was stated that
an increase in intensity produces an increase in per cent
correct recognition on articulation tests.

Improvement in intelligibility scores as a result
of intensity increase appears to be a diminishing function
as higher levels of intensity are reached, as illustrated
by the smaller difference in scores between 50 dB and 70
dB as compared to the difference in scores between 35 dB
and 50 dB. There is a rapid increase seen in per cent
correct recognition as intensity increases at lower levels
of intensity, but then there appears to be a tendency for
the articulation curve to level or assume a maximum score
at the higher intensity level without reaching one hundred
per cent correct recognition. If a curve of best fit were
drawn to produce such an articulation curve in Figure 2,
it would appear to be of a similar shape to those found
by Siegenthaler and Hardick56 in a study of intelligibility
scores for normal hearing adults and children under unfil-
tered conditions. The difference found between such a
curve based on the present study and compared to the Sie-
genthaler and Hardick study would be one of placement of

the curve rather than shape. That is, under the filtered

 

55French and Steinberg, op. cit., p. 117.

6Siegenthaler and Hardick, loo, cit.

47

lOO--

95‘

90" 480-cps 960 cps

85‘t
80--
751-
70..

65--

55-r- \ 7 \
505-
45..
40--
35..

Per Cent Correct

 

 

 

  

 

 

 

 

 

 

 

 

 

 

3335 \V \

re: .0002 dynes/cm2

Fig. 2.-—Mean Per Cent Scores Under Conditions of Filtering

48

conditions, the curve would be shifted to the higher inten-
sity levels, and depressed into the lower intelligibility
scores, but retain much the same shape.

Rejection of the second hypothesis indicates that
there are significant differences in intelligibility
scores obtained at one or more different intensity levels.
Figure 2 illustrates the mean per cent correct scores for
each bandwidth group at the three intensity levels used
in the present study. One can see from this chart that
there is a definite increase in intelligibility scores as
intensity is increased. However, the scores obtained are
lower than scores obtained from a similar articulation
test under unfiltered conditions and do not reach a level
of one hundred per cent correct recognition, regardless
of the amount of increase in intensity.

Analysis of the data obtained in this study re-
sulted in a failure to reject the hypothesis of no sig-
nificant interaction between bandwidth and intensity
levels. There are no outstanding scores obtained in
this study that would indicate that any particular band-
width reacts with a certain intensity level to produce

unusual results.

CHAPTER V

SUMMARY AND CONCLUSIONS

Summary
Intelligibility of speech has been studied under

many different conditions. This research has used test
material made up of phonetically balanced word lists,
nonsense syllables, continuous discourse, spondees, and
other types of speech material. Some studies have been
done employing normal hearing subjects and other studies
have utilized subjects having various types of hearing
impairments. Still other studies have attempted to simu-
late various types of hearing loss through the use of
low-pass, high-pass, and band-pass filters. Filters have
also been used to introduce frequency distortion into

the speech signal so as to examine the effects of such
distortion on intelligibility of speech. Filter systems
have been used singly or in combinations to investigate
the effects of a limited frequency spectrum upon intel-
ligibility. Many studies have reported the effects of
noise on intelligibility of speech, both in distortion-
free investigations and under conditions of various kinds

of distortion.

49

50

The present study investigated the effects of
band-pass filtering upon the intelligibility of a com-
mon monosyllable word list. It was asked whether there
would be a significant difference in intelligibility of
speech between two bands of frequencies which are of
different widths. A second aspect of this investigation
was concerned with the effects of variation in intensity
levels upon intelligibility. It was asked if there would
be an increase in the intelligibility of speech with an
increase in intensity level of the stimulus presentation.
Finally, this study searched for any effects of inter-
action between bandwidths and intensity levels.

Twenty-four normal hearing subjects served as
listeners in this study. These subjects were selected
from the graduate and undergraduate student population
of the Speech Department of Michigan State University.
No persons were used who had extensive acquaintance or
experience with the CID Auditory Test W-22. All sub-
jects demonstrated a hearing acuity level of 10 dB
(re: normal threshold) or better in both ears as mea-
sured by a portable pure tone audiometer in a sound
treated room.

The subjects were randomly divided into two groups
of twelve each. One group listened to the W-22 word
lists under a 480 cps bandwidth condition centered at

1560 cps. The second group of subjects listened to the

51

same lists under a 960 cps bandwidth condition at the same
center frequency. Both groups of subjects recorded re-
sponses to a practice list at 90 dB SPL before beginning
the actual experiment. Each group then listened to the
lists at three intensity levels (35 dB, 50 dB, 70 dB,

re: .0002 dynes per square centimeter). All listeners
recorded their responses to the test material on pre-
pared answer sheets.

Responses obtained from the two groups of subjects
were analyzed using a two-way mixed Type I design analysis
of variance. The result of this analysis indicated that
there was no significant difference in intelligibility
scores between a 480 cps bandwidth condition and a 960 cps
bandwidth condition, both of which are centered at 1560 cps.
The data showed that there was a significant difference in
intelligibility scores as a result of intensity changes.
There was found to be no significant interaction between
bandwidth and intensity, under the conditions of this

study.

Conclusions

 

From the findings of this study, it may be concluded
that:

1. Limiting the frequency spectrum through the
use of filtering produces a decrease in intelligibility

of common monosyllable words.

 

52

2. Increasing the width of a band of frequencies
from 480 cps to 960 cps while maintaining the same cen-
ter frequency of 1560 cps, does not significantly improve
intelligibility scores.

3. Under the conditions of this study, increasing
the intensity level of the stimulus presentation produces
an increase in the number of correct recognitions of the
CID Auditory Test W-22 word lists.

4. There was.found to be no interaction between
the bandwidths and intensity levels utilized in this

study.

Implications for Future Research

Analysis and discussion of the results of the
present study have led to the formulation of several
questions which have grown out of the data obtained in
this experiment.

It has been found that doubling the width of a
480 cps band of frequencies centered at 1560 cps did
not produce a significant difference in intelligibility
scores. This finding leads one to question how much
wider must the wide band be in order to obtain signi-
ficantly different intelligibility scores between the
two bandwidths.

Since there was no difference in intelligibility
scores between these two bandwidths at a center frequency

of 1560 cps, one wonders whether this would still be true

53

at other center frequencies, such as at a center fre-
quency of 1000 cps. Assuming that the factor of differ-
ence in bandwidth is not the same for other center fre-
quencies, another question is raised, namely, how much
difference is required for significant results at these
other levels?

These questions indicate merely a small portion of
the lack of knowledge regarding some areas of speech
intelligibility. Despite the abundance of research on
the intelligibility of speech, there is need of much

more information.

BIBLIOGRAPHY

Books

Davis, Hallowell. Hearing and Deafness. New York:
Holt Rinehart and Winsotn, Inc., 1963.

Fletcher, Harvey. Speech and Hearin in Communication.
New York: D. Van Nostrand Co. Inc., I953.

Gray, Giles W., and Wise, Claude M. The Bases of Speech.
New York: Harper and Brothers, 1959.

 

 

 

Hirsh, Ira J. The Measurement of Hearing. New York:
McGraw-HiII Book”Cb. Inc., 1952.

 

Lindquist, E. F. Design and Anal sis of Experiments in
Ps cholo yyand Education. oston: Houghton MiffIin
C00, 0

Newby, Hayes. Audiology. 2nd ed. revised. New York:
Appleton Century Crofts, 1964.

Stevens, 8. S. (ed.). Handbook of Ex erimental Psychology.
New York: John WiIey & Sons, Inc., 1951.

Travis, Lee Edward, et a1. Handbook of S eech Patholo .
New York: AppIeton Century Cro ts, Inc., 195 .

 

 

 

Periodicals

 

Black, John W. "Accompaniments of Word Familiarity,"
Journal of Speech and Hearing Disorders, 17 (1952),

Egan, James P. "Articulation Testing Methods," Laryngo-

Egan, James P. and Weiner, F. M. "On the Intelligibility
of Bands of Speech in Noise," Joupnal of the Acoustic
Society of America, 18 (1946), 435-441.

Elpern, Barry S. "Differences in Difficulty Among the CID

W—22 Auditory Tests," Laryngoscope, 70 (1957),
1560-1565.

54

 

55

Fletcher, H. and Steinberg, J. C. "Articulation Testing
Methods," Bell Systems Technical Journal, 8 (1929),

French, N. R. and Steinberg, J. C. "Factors Governing the
Intelligibility of Speech Sounds," Journal of the
Acoustical Society of America, 19 (I949?) 90-119.

 

Galt, Rogers G. "The Importance of Different Frequency
Regions for Speech Intelligibility," Journal of”
the Acoustical Society of America, 20 (1948), 592.

Giolas, Thomas G. and Epstein, Aubrey. "Comparative
Intelligibility of Word Lists and Continuous Dis-
course," Journal of Speech and Hearing Research,
6 (1963), 349-355.

Harris, J. Donald and Brown, Lawrence W. "Interactions
Among Bandwidth, Center Frequency, and Type of
Distortion in Speech Intelligibility," Journal
of the Acoustical Society of America, 34 (I962),
I999.

Harris, J. Donald, Harris, H. L., and Myers, C. K.
"The Importance of Hearing at 3KC for Understanding
Speeded Speech," Lagyngosc0pe, 70 (1957), 131-146.

 

Hawkins, J. E. and Stevens, S. S. "The Masking of Pure
Tones of Speech by White Noise," Journal of the
Acoustical Society of America, 22 (1950), 6-13.'

Hirsh, Ira J., et al. "Deve10pment of Materials for
Speech Audiometry," Journal of Speech and Hearing
Disorders, 17 (1952), 32I-337.

 

Hirsh, Ira J., Reynolds, E. G., and Joseph M. "Intelli-
gibility of Different Speech Materials," Journal
of the Acoustical Society of America, 26 (I954),
5304538.

Hudgins, C. V., et al. "The Development of Recorded
Auditory Tests for Measuring Hearing Loss for Speech,
Laryngoscgpe, 57 (1947), 57-89.

 

Huizing, H. S., Kruisinga, R. J., and Taseloar, M. "Trip-
lett Audiometry: An Analysis of Band Discrimination
in Speech Reception," Acta Oto-Larypgology, 51 (1960),

 

56

Jerger, James, et al.- "Some Relations Between Normal
Hearing for Pure Tones and Speech," Journal of
Speech and Hearing Research, 2 (1959), I26-I40.

Kryter, Karl D. "Effects of Ear Protective Devices on
the Intelligibility of Speech in Noise," Journal
of the Acoustical Societyypf America, 18 (I946),
413-417.

Kryter, Karl D. "Methods for the Calculation and Use of
the Articulation Index," Journal of the Acoustical
Society of America, 34 (1962), 1689-1692.

Kryter, Karl D. "Speech Bandwidth Compression Through
Spectrum Selection," Journal of the Acoustical
Society of America, 32 (1960), 547-556.

 

 

Licklider, J. C. R. "Effects of Amplitude Distortion
upon the Intelligibility of Speech," Journal of
the Acoustical Society of America, 18 (1946),
429-434}

 

Licklider, J. C. R. and Miller, G. A. "The Intelligi—
bility of Interrupted Speech," Journal of the
Acoustical Society of America, 20 (I948), 593.

Owens, Elmer. "Intelligibility of Words Varying in Famil-
iarity," Journal of §peech and HearingyResearch,
4 (1961), 113-129.

Peterson, G. and Subrahmanyan, D.‘ "Evaluation of Time-
Frequency Scanning for Narrowband Speech Trans-
mission," Journal of the Acoustic Society of
America, 31 (1959), 110-114.

Pollack, Irwin. "Effects of High Pass and Low Pass
Filtering on the Intelligibility of Speech in
Noise," Journal of theyAcoustical Society of.
America, 20 (1948), 259-266.

Shaw, W. A., Newman, E. R., and Hirsh, I. J. "The Dif-
ference Between Monaural and Binaural Thresholds,"
.Journal of Experimental Psychology, 37 (1947),
229-242.

Sigenthaler, Bruce B. and Hardick, Edward. "Intelligi-
bility Scores Using Various Phnetically Balanced
Word Lists," PennsylvaniaySpeech Annual, (1959).

 

57

Tiffany, William and Bennett, Delmond. "Intelligibility
of Slow-Played Speech," Journal of Speech and Hearing
Research, 4 (1961), 248-258. III

 

Unpublished Material

 

Hardick, Edward and Deal, Leo. "The Effects of Frequency
Distortion on Speech Discrimination," Unpublished
paper, Department of Speech, Michigan State Uni-
versity,‘l962.

 

58

 

more .mm new» .mm 36“ .mm one; .mm

 

mos .em mono .em Assess 36: .em mason .em
o>moe .mm rezone .mm Aneonc new .mm mam .mm
lessee memo .NN one: .NN memo .NN recess son .NN
looses omonn .Hm sou .AN ascou .HN Ame .Hm
Iowans ammo .om romeo mes .om umm .om Anemone muons .om
mo» .ma was .mH woe .ma m>nmo .ma
no» .me mad .me m>os .me roan .me
Hows .hH m>mn .ua pcmm .hH Annoy Ho .ha
m>6m .oe menu .oe . soc .oe meow .oa
monsoHo .mH ow .mH m>mo..ma Assoc macs .mH
Issac Hem .ee Hoo3 .ea use» .ea sew .eH
son .me sew .mH .muo .me won .me
88 .NH soon .NH sees .me pans .me
he .HH Amsmhv mm: .HH uHmEm .HH cop .HH
“some on .oe paw .oH awn .oe am .oe
cap .m SpooEm .m peso .o Ammmv mom .m
mocm .m Hflo .m coo .m Acnwv sumo .m
coo .5 sum .5 wmuu .n can» .h
camp .m co .m can .m Boa .m
cmE .m we .m scum .m mmmc .m
new .v uumum .v 38m .e mom .w
Aeneas canons» .m memo .m mass .m he pm
Aoesoso 6003 .N 63 .m Hes .N o>eo -m
lacunae peoso .H some pom .H page .H soap .H
as ewes mm omen em amen mm mmfle_

NNIB Emme NKOBHQD< .Q.H.U

d xHDmem<

59

um
“map
Ansocv s50
laaao as
Amway mop
Haas

use

xooo

we:

om

wwwum
mnoc3
cwco

0:3

umc

was

mus:
moms

Ens

>c3

coca

mo

mash
mHHOp
Amoco on

.om
.mv
.me
.hv
.ww
.mv
.vv
.mv
..Nv
.Hv
.ov
.mm
.mm
.hm
.mm
.mm
.vm
.mm
.Nm
.Hm
.om
.mm
.mm
.hN
.mm

ratio»: 4| ”9.1.2.... . f .. . .d. J 3r.

umm3
Aeroflot one
any

Hasc

use

NH

ones
Asocxv 0c
“flux
w>ocm
memo

map

cons

mum

so»

Acapv macp
mmzo

on

Show
and:
m>oau
Mom
Hflmso
HHflQ
sonny

.om
.mv
.mv
.hv
.mv
.mv
.vw
.mw
.Nv
.Hv
.ov
.mm
.mm
.hm
.mm
.mm
.vm
.mm
.mm
.Hm
.om
.mm
.mm
.hm
.om

mmmm
ucam

A0“ rosuu co»
Aupomv duo»
mmOp

swam
acvamzv mmz
ans:

Aummpv mum»
loses Ham
moo

305m

ummco

Issac an
pan

Anum>v m>fla
mono»

xmo

hmx

Acmmcv GHQ
czo

coca

Acosv mco
loses map
HHHB

.om
.mw
Paw
«we
.mv
.mv
pvv
.mv
.Nv
.Hv
.ov
.mm
.mm
.bm
.mm
.mm
.wm
.mm
.mm
.Hm
.om
.mm
.mm
.mm
.mm

Amamfimv mama
fiasco
nos
08
undo
mwzmv 50%
Bus
mmmcx
pass
up
cflxm
ocumn
30H
uHmm
w>oum
msuu
mm
30:0
can
oases
EH:
Hoom
mcflsp
ms
H30

.om
.mv
«we
.hv
.mw
.mv
.vv
.mv
.Nv
.Hw
.ov
.mm
.mm
.bm
.mm
.mm
.wm
.mm
.mm
.Hm
.om
.mm
.mm
.hm
.om

17.
18.
19.
20.
21.
22.
23.
24.

25.

Subjects Answer Sheet

APPENDIX B

 

26.

 

27.

 

28.

 

29.

 

30.

 

31.

 

32.

 

33.

 

34.

 

35.

 

36.

 

37.

 

38.

 

39.

 

40.

 

41.

 

42.

 

43.

 

44.

 

45.

 

46.

 

47.

 

48.

 

49.

 

50.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

‘l. 26.
2. 27.
3. 28.
4. 29.
5. 3o.
- 6. 31.
7. 32.
8. 33.
9. 34.
filo. 35.
ll. 36.
12. 37.
13. 38.
14. 39.
15. 4o.
16. 41.
17. 42.
18. 43.
19. 44.
20. 45.
21. 46.
22. 47.
23. 48.
p 24. 49.
25. 50.

 

6O

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

§ 3 i ‘fln’n!

 

8‘11

APPENDIX C

You are going to hear four lists of one-syllable
words. Some lists will be very loud and some lists
will be very quiet. Each list contains fifty words which
correspond with the fifty numbered blank spaces on the
answer sheet in front of you. You are to write the word
you hear, or think you hear, in the apprOpriate numbered

space. If you miss a word or cannot make out what the word

 

is, leave the space blank-and go on to the next one. List-
en carefully and write down as many words as you can.

At the end of each list will be a short pause during
which you should locate the next list of blanks on your
answer sheet. You will notice that the lists on your an-
swer sheet are numbered from one to four. Be sure to use
them in that order.

Please work as quietly as possible in order to re-
duce the amount of noise in the room.

Thank you for your assistance.

61

 

mmull")![(1)1111

0