J Am Acad Audiol 2 :91-98 (1991)
Perceptual Features for Normal Listeners'
Phoneme Recognition in a Reverberant
Lecture Hall
Jeffrey L. Danhauer*
Carole E . Johnsont
Abstract
Johnson et al (1988) found that normal-hearing listeners' consonant
recognition scores
on a nonsense syllable test were significantly better when they were
seated within versus
beyond the critical distance of a typical lecture hall . The present study
used a Sequential
Information Analysis (SINFA) to document a priori perceptual features for
Johnson et al's
subjects' confusion errors to the initial and medial consonant portions of
the stimuli . The
features and the percent of information transmitted for them differed as
a function of consonant position in the stimulus and whether the subjects were seated within
or beyond
the critical distance of the room . Generally, more information was transmitted
within rather
than beyond the critical distance for both initial and medial consonants .
However, less information was transmitted for features for the medial than for the initial
consonant position . Beyond the critical distance, high-frequency features (e .g ., sibilant) were
transmitted
somewhat better than those based more on low-frequency energy (e .g .,
voicing) . Place
of articulation features were not transmitted as well for the initial as for the
medial position
and they suffered in both positions as distance increased. The effects of
reverberation on
the subjects' perception are discussed and compared to results of other
studies.
Key Words: Consonant recognition, critical distance, information transmission,
lecture hall,
normal-hearing, perceptual features, reverberation
his study evaluated perceptual features
for normal-hearing listeners' consonant
T recognition in a typical reverberant lecture hall. Several studies (e.g., Finitzo-Hieber
and Tillman, 1978 ; Gelfand and Silman, 1979 ;
Nabelek and Mason, 1981) have found that
reverberation adversely affects listeners' speech
recognition abilities. These studies have also
found that the longer the reverberation time of
a room, the more adverse are the effects on
speech understanding and that consonants in
'Department of Speech and Hearing Sciences, University of California Santa Barbara, Santa Barbara, California
tDepartment of Communicative Disorders, University of
Oklahoma Health Sciences Center, Oklahoma City, Oklahoma
Reprint requests : Jeffrey L . Danhauer, Department of
Speech and Hearing Sciences, Snidecor Hall, University of
California Santa Barbara, Santa Barbara, CA 93106
the initial position of words are less affected by
reverberation than those appearing in the middle or at the end of words.
The effects of reverberation also depend on
speaker-to-listener distance. Peutz (1971) investigated the effects of loudspeaker-to-listener distance on the phoneme recognition abilities of
normal-hearing listeners in rooms differing in
volume and reverberation time. He found that
while speech understanding was essentially
unaffected in front of a loudspeaker, it gradually decreased with increasing distance from the
source up to the critical distance (point where
direct and reflected energy is equal) of the room.
However, beyond the critical distance, speech
recognition remained essentially unchanged
with further increases in loudspeaker-to-listener
distance.
Johnson et al, (1988) evaluated normal-hearing listeners' nonsense syllable test (Edgerton
Journal of the American Academy of Audiology/Volume 2, Number 2, April 1991
and Danhauer, 1979) scores in a reverberant lecture hall having a reverberation time of 1 .9 s.
They found that vowel scores (M = 89.9%) were
not significantly affected by distance. However,
consonant scores (M = 61 .8%) were significantly
better (68% correct) within the critical distance
of the room than at two distances (59.5% correct) that were beyond the critical distance. The
scores beyond the critical distance were not significantly different from each other.
The findings of Peutz (1971) and Johnson
et al (1988) demonstrated that beyond the critical distance, the full effect of masking by reverberation occurs and remains constant independent of loudspeaker-to-listener distance . However, even within the critical distance, consonant
recognition was difficult for normal-hearing
listeners. Edgerton and Danhauer (1979) reported that the mean correct consonant score on
these stimuli for normal-hearing subjects (listening diotically under earphones at 55 dB sensation level re SRT) was only 85 .8 percent (SD =
10 .1).
An analysis of perceptual features from subjects' consonant confusion errors (Johnson et al,
1988) may provide insight as to why their consonant recognition performance was poor in
these conditions . Related studies (Miller and
Nicely, 1955 ; Wang and Bilger, 1973 ; Gelfand
and Silman, 1979) have used information transmission analyses to evaluate sets of a priori perceptual features for normal-hearing subjects'
consonant confusions in filtering, noise, and/or
reverberation. Other studies (e.g., Doyle et al,
1981 ; Danhauer et al, 1986) have used a multidimensional scaling procedure to analyze a
posteriori features for normal-hearing listeners'
initial and medial consonant confusion errors to
nonsense syllable stimuli.
There is a major difference between the a
posteriori and the a priori types of analyses. In
the a posteriori methods, a set of features is
derived from the data via empirically predicted
relationships among the stimulus items. However, considerable interpretation of the results
is usually required on the part of the experimenters. The a priori methods determine the
subjects' levels of performance (amount of transmitted information) for a set of predetermined
features, which the a posteriori methods cannot
do. Thus, the a posteriori methods are useful in
arriving at a set of features for a particular body
of data when the features for it are not known
in advance. However, the a priori methods are
more useful in determining subjects' performance on specific features once the set is known.
We may assume that the set of features has
been identified for the stimuli used in this study
(e .g., Doyle et al, 1981), and that it is appropriate to use the a priori method to evaluate the
effects of reverberation on these normal-hearing
listeners' perception of certain features . Thus,
the present study analyzed subjects' consonant
confusion errors (Johnson et al, 1988) using an
a priori. method to determine the effects of
loudspeaker-to-listener distance (i .e ., within
versus beyond the critical distance of the room)
on their perceptual features for initial and
medial consonants .
METHOD
Subjects
The subjects were 12 women between 19
and 35 (M = 25) years of age. They were all
speech and hearing majors in college and had
normal hearing (pure-tone thresholds of 15 dB
HTL or better for the octave frequencies between 250 and 8000 Hz and no history of hearing problems).
Stimuli
The stimulus was an Auditec of St. Louis
audio cassette tape recording of List A of the
Edgerton and Danhauer (1979) Nonsense Syllable 'Ibst (NST) as shown in Figure 1. The
NST is a 25-item consonant-vowel-consonantvowel (CVCV) test of phoneme recognition. Six
randomizations of List A are available on the
tape. Edgerton et al (1981) found no practice effects for the NST when up to seven administrations of List A were presented to normal-hearing
listeners in a single test session. The NST items
are presented in an open-response format following the carrier word "say." The subject's task is
to repeat what is heard while the examiner
phonemically transcribes any errors to the
stimuli on special response forms. There are 50
consonants (25 initial and 25 medial) and 50
vowels (25 medial and 25 final) on NST List A.
Test Environment
Each subject was tested individually in a
college lecture hall that contained furniture, the
Reverberant Room Features/Danhauer and Johnson
FORM
A
1 a~j o
14
C)
15
2 QI
3
b~ ,~O
4 fLJel
16
17
5
fIe~0
18
6
f*1Sa2
19
7 VUVI
20
s
d3 afe
b uma
10
S
11
12
JE
feed6 T [~e
13 bntI
21
22
23
ba,na
baebae
bUmas
f i t fa
Pnd U
tfo e a-0Pits
d3AbU
eA f I
habo
24
daea
25
ni9U
subject, experimenters, and equipment. The
room had a volume (V) of 415 m3 and a reverberation time (T) of 1 .9 s. Using a sound absorption value of 0.2, the critical distance (CD) of
the room was calculated to be 3 m according to
Peutz's (1971) equation : CD = 0.2
(V/T).
Subjects were tested at three loudspeaker-tolistener distances : at 1 m (within the critical distance of the room) and at 4 m and at 8 m (both
beyond the critical distance).
The reverberation time for the room was
measured by firing a cap pistol in one corner of
the room and recording it with a high quality
reel-to-reel tape recorder (Sony APR-2003) positioned in the opposite diagonal corner. The output of the tape recorder was connected to a
one-third octave band filter that was connected
to a graphic level recorder (Bruel & Kjaer 2203).
Decay of the impulse noise was measured as a
function of time using a reverberation time protractor (Bruel & Kjaer SC2361) at a paper speed
of 100 mm/s . Reverberation times for the onethird octave bands centered at 250, 500, 1000,
2000, 4000, and 8000 Hz were 1 .9 s, 2.0 s, 1 .9
s, 1.8 s, 1.4 s, and .9 s, respectively. The average
reverberation time (500, 1000, and 2000 Hz onethird octave bands) was 1 .9 s.
The NST stimuli were presented from a tape
player through a loudspeaker to the subject who
sat in a desk facing (at zero degrees azimuth)
the loudspeaker. The stimuli peaked at about 75
dB A on a sound level meter at the 1 m distance
and at about 71 dB A at the 4 m and 8 m distances. The ambient noise in the room peaked
at about 54 dB A. Thus, the signal-to-noise ratio (S/N) ranged from about +21 to +17 dB .
Danhauer, Doyle, and Lucks (1986) found no significant difference between +20 and +10 dB
S/Ns for normal listeners' performance on the
NST. Thus, the difference for the within versus
beyond the critical distance conditions should
not have been a factor in this study.
Measurements were also made of ambient
noise levels for octave-band filters with center
frequencies from 250 to 8000 Hz using a precision sound level meter (Bruel & Kjaer 2203) and
a slow meter setting. Measurements made at all
three test locations were within ±1 dB of each
other. Ambient noise measurements (i.e .,
without the NST signal present) were: 58 dB
(250 Hz), 55 dB (500 Hz), 47 dB (1000 Hz), 40
dB (2000 Hz), 35 dB (4000 Hz), and 29 .5 dB
(8000 Hz). Measurements made with the NST
signal present for the 4 m and 8 m locations were
the same: 68 dB (250 Hz), 74 dB (500 Hz), 66
dB (1000 Hz), 56 dB (2000 Hz), 42 dB (4000 Hz),
and 29 dB (8000 Hz). Thus, the resulting S/Ns
of +10 dB (250 Hz), +19 dB (500 Hz), +19 dB
(1000 Hz), +16 dB (2000 Hz), +7 dB (4000 Hz),
and -0 .5 dB (8000 Hz) were most favorable in
the lower frequency bands .
Procedure
The subjects were told that they would hear
nonsense syllables, to repeat them aloud, and
guess if unsure of a response . Each subject
received one randomized form (from the six
available) of NST List A at each of the three distances. The randomizations and distances were
counterbalanced across subjects . The entire
procedure required approximately 20 min
per subject.
Two experimenters experienced with the
NST and phoneme transcription sat 2 m from
and facing perpendicular to the listener. One experimenter sat on each side of the subject so
that visual cues were available to them in tran-
Journal of the American Academy of Audiology/Volume 2, Number 2, April 1991
scribing responses. The 2 m subject-toexperimenter distance was selected so that the
experimenters would be within the critical distance of the room to see and hear the subject
clearly, but not be so close that they would interfere with the room acoustics. The subjects all
repeated their responses so that they were loud
and clear to the experimenters. Also, the use of
visual cues by the experimenters in transcribing the subjects' responses should have compensated for any problems due to subjectto-experimenter distance. Both experimenters
were also present in the room while all the reverberation measurements were made to account
for their presence during the test sessions . Thus,
because the experimenters were present at all
three test locations, their effects on room acoustics should not have been a factor in the subjects' performance at the three distances.
The experimenters independently phonemically transcribed each subject's responses to the
stimuli. Inter-judge reliability was 96 percent.
Phoneme scoring was used in which each consonant or vowel recognized correctly in each of
the 25 CVCV items was awarded 1 percent
toward a possible 100 percent score. Separate
consonant scores and vowel scores were also
derived by awarding 2 percent for each consonant or vowel component of each item recognized correctly.
Data Preparation
The subjects' consonant recognition scores
were significantly better when they were within
(1 m) versus beyond (4 m and 8 m) the critical
Table 1
distance for the room . However, the scores for
the 4 m and 8 m locations were not significantly different from each other. Therefore, the subjects' errors were converted to two separate
confusion matrices (one for the 1 m location
within and one pooling both the 4 m and 8 m
locations beyond the critical distance). Further,
because the stimuli comprising the initial and
medial consonants of the NST differ, separate
confusion matrices were also constructed for
each consonant position .
Inspection of each subject's confusion matrices produced no apparent individual differences across the subjects. Thus, all the subjects'
data were pooled for each of the conditions noted above. This resulted in 300 observations per
matrix (25 consonants X 12 subjects X 1 distance, 1 m) for the within condition and 600 (25
consonants X 12 subjects X 2 distances, 4 m
and 8 m) for the beyond, which are fewer than
those used in some other studies (e.g., Miller and
Nicely, 1955 ; Wang and Bilger, 1973 ; Gelfand
and Silman, 1979). The matrices are not provided here, but are available from the authors.
The subjects' consonant confusion matrices
were submitted to a Sequential Information
Analysis (SINFA) program (Wang and Bilger,
1973) to determine the amounts of information
transmitted for selected a priori perceptual features. SINFA performed a multivariate uncertainty analysis on the confusion matrices . The
features used in this analysis included those derived from earlier a posteriori analyses of other
normal-hearing listeners' responses to consonants of the NST (Danhauer et al, 1986 ; Doyle
et al, 1981) . The features used in the present
analyses are shown in Table 1 for the initial and
the medial consonants .
Features Used in the SINFA for the Initial and Medial Consonants
Medial Consonants
Initial Consonants
Feature
Uncond. Info. '
Feature
Uncond . Info .
Stop
943
Dental
Affricative
.971
529
971
904
401
528
794
942
401
Fricative
Front/back
Voiceless-stop
Nasal
Sonorant
Dental
Sibilant
Fricative
Front/back
Voicing
Nasal
Sibilant
Voicing
242
855
971
'Unconditional information is shown in bits .
903
528
999
Reverberant Room Features/Danhauer and Johnson
Table 2
Summary of SINFA for Initial Consonants
Beyond Critical Distance
Within Critical Distance
Feature
Ranks
Info . Transmitted
(In %)
(In Bits)
Sibilant
Affricative
Voicing
Stop
Dental
Front/back
Nasal
Fricative
Tot. bits info . tra ns .
For matrix =
Obtained =
Prop , info . trans.
Info . Transmitted
(In Bits)
(In %)
Feature
Ranks
829
97
631
88
Sibilant
Affricative
Nasal
59
53
Voicing
Front/back
478
.171
Stop
147
254
491
191
161
037
.037
2 .904
2 .566
92
76
76
53
Fricative
Dental
793
206
106
93
75
73
197
26
131
67
42
26
25
2 .323
1 .902
88
RESULTS
B
riefly, the SINFA program used in the
present analyses first estimated the unconditional transmitted information for each feature in bits. As seen in Table 1, the stimulus
information for each feature varied considerably across the features . The percentage of information transmitted for each feature is also
shown and was used to normalize the features
for inequalities in stimulus feature information.
Thus, the subjects' relative performance on the
features is presented as the percentage of information transmitted for each feature.
The program then partialed out the effects
of one feature on another by completing a series of iterations in which the percent of conditional information transmitted for each of the
features was determined. The feature identified
as having the highest performance on iteration
82
one (i .e ., the unconditional transmitted information discussed above) was partialed out for iteration two and the conditional information for the
remaining features was determined independent
of that for the feature identified in iteration one.
That feature identified as having the highest
performance in iteration two was then also partialed out for iteration three, and so on . This
process continued to partial out, in each subsequent iteration, the effects of features identified
in previous iterations until all the features were
accounted for. For a complete discussion of the
SINFA procedure, see Wang and Bilger (1973) .
The results of the SINFA are summarized
for each distance in Tables 2 and 3 for the initial and medial consonants, respectively. The tables show a ranking of the features in bits and
according to the percent of information transmitted for each. The tables also show the total
information transmitted for each matrix and the
Table 3 Summary of SINFA for Medial Consonants
Beyond Critical Distance
Within Critical Distance
Feature
Ranks
Info. Transmitted
(In Bits)
(In %)
Feature
Ranks
Info . Transmitted
(In Bits)
(In %)
Voicing
Voiceless-stop
Front/back
Nasal
928
.282
347
320
121
93
91
88
86
65
Voiceless-stop
Sibilant
Front/back
Nasal
313
388
319
090
89
88
76
61
111
111
65
65
Dental
Fricative
Voicing
243
243
113
40
40
26
Sibilant
Sonorant
Dental
Fricative
Tot. bits info . trans .
For matrix =
Obtained =
Prop . Info . trans . =
111
2 .443
2 .157
65
88
Sonorant
243
2 .165
1 .566
40
72
Journal of the American Academy of Audiology/Volume 2, Number 2, April 1991
amount of information obtained in bits, and the
proportion of information transmitted in percent . The results on all three categories were
better when the subjects were within rather
than beyond the critical distance . This was true
for both initial and medial consonants, although
the scores for the medial consonants were somewhat lower.
Sibilant and affricative features were transmitted well at both distances for the initial consonants. Interestingly, sibilant was only
transmitted well beyond the critical distance for
the medial consonants. Voicing was transmitted
much better within the critical distance than beyond it for both consonant positions. With the
exception of front/back in the medial position,
features relating to place of articulation were not
transmitted very well in either position for
either distance .
Tables 2 and 3 show that over 50 percent of
the information was transmitted for each of the
features analyzed for the initial and medial consonants when presented within the critical distance. Considerably less information was
transmitted for most of the features when
presented beyond the critical distance. Thus, the
alterations in the rankings of the features noted across distances and consonant positions relate to the reverberant characteristics of this
particular room and to whether or not a vowel
preceded the consonant in the stimulus .
DISCUSSION
Effects of Loudspeaker-to-Listener Distances
and Noise on Features
The rankings and percentages of information transmitted for most of the features in this
analysis were consistent with the results of Gelfand and Silman (1979) who evaluated normalhearing listeners' recognition of consonants in
monosyllabic words recorded in a small room
having a 0.8 s reverberation time . For example,
the sibilant feature (characterized by highfrequency energy and present in the phonemes
/s, z, J , ,t J , d3 /) was transmitted well in all
conditions . Gelfand and Silman (1979) found
that the information transmitted for sibilant
was barely affected by reverberation.
Our sibilant feature is also similar to Miller
and Nicely's (1955) duration feature that included the phonemes /s, z, (, 3 /. Their duration
feature was salient for their normal-hearing lis-
teners only in relatively quiet conditions or
when the stimuli were presented through highpass filters; it was perceived poorly in conditions
of noise or low-pass filtering. However, Miller
and Nicely (1955) used much poorer S/Ns and
their white noise probably had different spectral characteristics than the ambient noise in
our study. While white noise has equal intensity at all frequencies, the spectrum of the ambient noise in our lecture hall had greater
amplitude in the lower frequencies than in the
higher frequencies .
Our results for affricative (Its and d3 /),
another high-frequency feature that is somewhat redundant with the sibilant feature,
showed that it was transmitted well at both distances for the initial consonants . This feature
was not analyzed for the medial consonants because /v and d3 / do not occur in that position
of the NST stimuli. Inspection of Gelfand and
Silman's (1979) matrices also reveals that all of
their subjects always perceived /tV' and d3 / correctly whether in quiet or under reverberation
and regardless of consonant position in the
stimulus. Although affricative only ranked third
overall of the five features analyzed by Miller
and Nicely (1955), they included the fricative
phonemes in their affricative category. Interestingly, the fricative feature was also not transmitted well in any conditions of the present
study nor in that of Gelfand and Silman. Thus,
the results of these studies are in fair agreement.
These findings probably relate to the lowfrequency reverberant characteristics of our particular lecture hall that may have masked many
of the cues available to the listeners when they
were situated beyond the critical distance. The
1.9 s reverberation time for this room was considerably longer than that for typical living
rooms or small offices having reverberation
times between 0.4 and 0.8 s (Nabelek and Nabelek, 1985). Normal-hearing listeners can tolerate up to about 1 .2 s of reverberation in
favorable S/Ns without speech recognition being adversely affected (Nabelek, 1980).
Our 1 .9 s reverberation time is similar to
that found in small auditoriums and churches
(Nabelek and Nabelek, 1985). The 1.9 s time also
falls between two (1 .6 s and 2.3 s) of the four
times that Moncur and Dirks (1967) used to
evaluate the effects of reverberation on binaural
speech intelligibility. They considered these
times to be relatively long and they produced
word recognition scores between approximately 67 and 60 percent, respectively.
Reverberant Room Features/Danhauer and Johnson
Thus, as would be predicted by Moncur and
Dirks (1967), our subjects had difficulty perceiving the stimuli within and beyond the critical
distance of the lecture hall. This was reflected
both in their percent correct scores and in the
information transmitted by the perceptual features. Features characterized mainly by highfrequency energy (e.g., sibilant and affricative)
were less affected by the predominantly lowfrequency reverberant noise in the room than
were features that are more low-frequency based
(e.g., voicing and nasality) . For example, voicing
was transmitted better within than beyond the
critical distance. The finding that voicing,
sonorant, and nasal were not transmitted as well
as some of the other features was somewhat unexpected based on Miller and Nicely's (1955)
data that showed that these features were
among the most salient in their various noise
and filtering conditions.
Features relating to place of articulation
were transmitted somewhat better than expected from Gelfand and Silman's (1979) results
which showed that place was degraded more
than any other feature in reverberation. Our
place features were generally perceived better
within than beyond the critical distance. This
follows from Miller and Nicely (1955) who noted
that those sounds that are easiest to see are
often hardest to hear.
Initial Versus Medial Feature Differences
Some interesting differences were observed
among the features for the two consonant positions of the NST They employ different stimulus sets, and the medial consonants are preceded
by a vowel while the initial consonants are not.
Generally, the features for the initial consonants
had higher percentages of information transmitted than those for the medial consonants . As
seen in Tables 2 and 3, the total information
transmitted in bits was also greater for the initial (2 .566 within and 1.902 beyond the critical
distance) than for the medial (2 .157 within and
1.566 beyond the critical distance) consonants.
Further, the proportion of information transmitted for the confusion matrices was: (a) initial
consonants within the critical distance, 88 percent ; (b) initial-beyond, 82 percent; (c) medialwithin, 88 percent ; and (d) medial-beyond, 72
percent. These results are in fair agreement with
the subjects' mean percent recognition scores
and seem to indicate that the initial consonants
were less affected by the reverberation of the
room than those in the medial position. These
results are in agreement with those of Gelfand
and Silman (1979) and Nabelek and Mason
(1981) who found that initial consonants were
less affected by reverberation than final consonants .
Nabelek and Nabelek (1985) noted that
reverberation causes sounds to overlap one
another such that the prolonged acoustic energy of vowels masks consonants following vowels
more than those preceding vowels . Thus, our
results make sense considering that the medial
consonants of the CVCV stimuli could have been
masked by the prolongation of the preceding
vowel due to reverberation, while the initial consonants were not influenced by a preceding
vowel.
Our findings regarding the importance of
certain features relative to the position of the
consonant within the nonsense syllable are also
in agreement with those of Wang and Bilger
(1973) . They tested listeners with CUs and VCs
in quiet and at six S/Ns from -10 dB to +15
dB at 5 dB intervals. Wang and Bilger found
that voicing, fricative, sibilant, duration, and
place (features from their study that were common to ours) were all perceived better in CUs
than in VCs in noise. They found that voicing
was transmitted best in noise (at all S/Ns except
+15 dB), but as the S/N improved, the differences in the information transmitted by the features in each position began to diminish. The
differences were negligible in quiet, except that
the importance of duration, sibilant, fricative,
and place switched (i .e., they had more information transmitted in the VC position than in the
CV). Unfortunately, both reverberation and
noise are present in real classroom listening
situations and interact synergistically to degrade speech recognition to a degree that can
not be predicted by summing the effects of each
factor separately (Finitzo-Hieber and Tillman,
1978 ; Nabelek and Mason, 1981).
Another difference noted between the
results for the two consonant positions related
to the relative order of the feature rankings according to the percentage of information transmitted. The order was fairly well maintained for
the initial consonants at both distances
(sibilant, affricative, and voicing were among the
first four features for each distance). However,
there were several alterations in the feature rankings for the medial consonants across the two
Journal of the American Academy of Audiology/Volume 2, Number 2, April 1991
distances. For example, voicing (93%) and
sibilant (65%) were ranked first and fifth, respectively within the critical distance and last (voicing 26%) and first (sibilant 88%), respectively
beyond the critical distance.
For the medial consonants, the voicelessstop feature (which only includes two phonemes,
/t and tf / and is also characterized by highfrequency energy) was perceived well at both distances suggesting again that the room acoustics had less effect on high-frequency sounds
than on low-frequency sounds. As noted earlier,
the S/N was more favorable in the lower than in
the higher octave bands and the reverberation
time was less favorable for the lower frequency
one-third octave bands. Thus, because the lowfrequency features were more affected by reverberation than the higher-frequency features, it
is doubtful that the S/N caused the reduction
in the percentage of information transmitted for
medial consonant features beyond the critical
distance.
SUMMARY
T
his study offers insight about the subjects'
perception of the consonants of the NST
when presented in a reverberant lecture hall.
These results go beyond the information provided by the simple percent correct consonant
recognition scores that showed that the subjects' performance was poorer when they were
situated beyond rather than within the critical
distance of the room, but failed to demonstrate
how they were poorer. This analysis of the subjects' error data determined their performance
for each of the perceptual features investigated . Generally, the subjects performed better
within than beyond the critical distance of the
room, but alterations were noted for the salience
of certain features depending on whether the
consonants were in the initial or medial position
of the stimuli. The fact that high-frequency features (e.g., sibilant) were transmitted somewhat
better than those based more on low-frequency
energy (e.g., voicing), suggests that the reverberant characteristics of the room affected the
subjects' perception of the features. These results help show why preferential seating aids the
perception of certain aspects of a stimulus and
not others . Our results are in agreement with
other studies that have used more observations
and different stimuli. Further research is war-
ranted to substantiate the stability of our findings with a larger number of observations .
Acknowledgments. Part of this research was supported
by a University of California General Research Grant
(!l 8587529-19900-7) .
REFERENCES
Danhauer JL, Abdala C, Johnson CE, Asp CW (1986).
Perceptual features from normal-hearing and hearingimpaired children's errors on the NST Ear Hear
7:318-322 .
Danhauer JL, Doyle PC, Lucks LE . (1986) . Effects of
signal-to-noise ratio on the Nonsense Syllable Tbst. Ear
Hear 7 :323-324 .
Doyle KJ, Danhauer JL, Edgerton BJ . (1981) . Features
from normal and sensorineural listeners' nonsense
syllable test errors. Ear Hear 2:117-121 .
Edgerton BJ, Danhauer JL . (1979). Clinical Implications
of Speech Discrimination Testing Using Nonsense Stimuli. Baltimore: University Park Press.
Edgerton BJ, Danhauer JL, Rizzo S. (1981). Practice effects for normal listeners' performance on a nonsense
syllable test. J Aud Res 21 :125-131 .
Finitzo-Hieber T, Tillman TW (1978) . Room acoustics effects on monosyllabic word discrimination ability for normal and hearing-impaired children . J Speech Hear Res
21 :440-458 .
Gelfand SA, Silman S. (1979) . Effects of small room reverberation upon the recognition of some consonant features.
J Acoust Soc Am 66 :22-29 .
Johnson CE, Nabelek AK, Asp CW, Danhauer JL. (1988) .
Loudspeaker- to-listener distance and phoneme recognition in a reverberant lecture hall. Paper presented at the
Annual Convention of the American Speech-LanguageHearing Association, Boston.
Miller G, Nicely PE . (1955) . An analysis of perceptual
confusions among English consonants . JAcoust Soc Am
27:338-352 .
Moncur JP, Dirks D. (1967). Binaural and monaural
speech intelligibility in reverberation . JSpeech Hear Res
10:186-195.
Nabelek AK . (1980) . Effects of room acoustics on speech
perception through hearing aids . In: Libby R,
ed . Binaural Hearing and Amplification. Chicago: Zenetron Inc.
Nabelek AK, Mason D . (1981) . Effect of noise and reverberation on binaural and monaural word identification
by subjects with various audiograms . JSpeech HearRes
24:375-383 .
Nabelek AK, Nabelek IV (1985). Room acoustics and
speech perception . In : Katz J, ed . Clinical Audiology. 3rd
ed. Baltimore: Williams & Wilkins, 834-846.
Peutz VMA . (1971). Articulation loss of consonants as
a criterion for speech transmission in a room . J Audio
Eng Soc 19 :915-919 .
Wang MD, Bilger RC. (1973) . Consonant confusions in
noise: a study of perceptual features. J Acoust Soc Am
54:1248-1266.