Download frequency discrimination ability and stop

Journal of Speech and Hearing Research, Volume 32, 133-142, March 1989
FREQUENCY DISCRIMINATION
ABILITY AND STOPCONSONANT IDENTIFICATION
IN N O R M A L L Y H E A R I N G A N D
HEARING-IMPAIRED
SUBJECTS
MARLEEN T. OCHS*
Radford University
LARRY E. HUMES
Indiana University
RALPH N. OHDE
D. WESLEY GRANTHAM
Vanderbilt University
Identification of place of articulation in the synthesized syllables/bi/,/di/, and/9i/was examined in three groups of listeners: (a)
normal hearers, (b) subjects with high-frequency sensorineural hearing loss, and (c) normally hearing subjects listening in noise.
Stimuli with an appropriate second formant (F2) transition (moving-F2 stimuli) were compared with stimuli in which F2 was
constant (straight-F2 stimuli) to examine the importance of the F2 transition in stop-consonant perception. For straight-F2 stimuli,
burst spectrum and F2 frequency were appropriate for the syllable involved. Syllable duration also was a variable, with formant
durations of 10, 19, 28, and 44 ms employed. All subjects' identification performance improved as stimulus duration increased.
The groups were equivalent in terms of their identification of/di/ and /gi/ syllables, whereas the hearing-impaired and
noise-masked normal listeners showed impaired performance for/bi/, particularly for the straight-F2 version. No difference in
performance among groups was seen for/di/and/gi/stimuli for moving-F2 and straight-F2 versions. Second-formant frequency
discrimination measures suggested that subjects' discrimination abilities were not acute enough to take advantage of the formant
transition in the/di/and/9i/stimuli.
The speech perception abilities of hearing-impaired
subjects have commonly been examined by organizing
and describing consonant recognition performance according to a distinctive feature system (Bilger & Wang, 1976;
Danhauer & Lawarre, 1979; Gordon-Salant & Wightman,
1983; Owens, Benedict, & Schubert, 1972; Sher & Owens,
1974; Walden & Montgomery, 1975; Wang, Reed, & Bilger,
1978). This approach represents an important first step in
our understanding of speech perception in the hearing
impaired. Interpretation is limited, however, because the
data describe the types of errors made but provide little
information about the underlying acoustic/perceptual
bases for the errors. Another approach to the examination
of speech perception in the hearing impaired has been to
establish first which acoustic characteristics of phonemes
are necessary for phoneme identification, and then to
determine which psychoacoustic abilities are related to
the perception of such cues (e.g., Buus, Miller, & Scharf,
1983; Dreschler & Plomp, 1985; Turner & Van Tasell,
1984; Tyler, Summerfield, Wood, & Fernandes, 1982).
In the present study, the perception of place of articulation for stop consonants is examined for two reasons.
First, the most common perceptual errors made by the
hearing impaired involve place of consonant articulation
(Byers, 1973; Owens et al., 1972; Owens & Schubert,
1968; Reed, 1975; Sher & Owens, 1974). Second, recent
studies have defined several of the acoustic characteris-
tics important for identification of stop-consonant place of
articulation in normally hearing subjects. There is evidence that context-invariant acoustic correlates of stopconsonant place of articulation can be found in the first 20
to 40 ms of a CV syllable (Blumstein & Stevens, 1979;
Kewley-Port, 1983). Stevens and Blumstein (1978) hypothesized that the auditory system integrates spectral
energy at syllable onset to produce a static representation
of place of articulation for stop consonants, and that the
listener is able to categorize the spectrum generated at
syllable onset according to a set of templates for the
various places of articulation. Their data suggested, moreover, that dynamic formant-transition information is not
necessary for accurate identification of voiced stop consonants. When stop-consonant syllables were manipulated to keep the frequency of the second and thirdformant transitions constant [called straight transition
stimuli by Blumstein & Stevens (1980) and Stevens &
Blumstein (1978)], identification performance did not
deteriorate substantially (Blumstein & Stevens, 1980).
In contrast, there are data that support the perceptual
importance of time-varying 'acoustic characteristics at syllable onset (Dorman, Studdert-Kennedy, & Raphael, 1977;
Hannley & Dorman, 1983; Kewley-Port, 1983; KewleyPort, Pisoni, & Studdert-Kennedy, 1983; Lahiri, GewilMa, &
Blumstein, 1984; Walley & Carrell, 1983). Kewley-Port et
al. have proposed that amplitude spectra are updated repeatedly in the auditory system during the course of the first
20 to 40 ms of the syllable and that this dynamic information
is important for identification of place of articulation. They
*Currently affiliated with Vanderbilt University.
© 1989, American Speech-Language-Hearing Association
133
0022-4685/89/3201-0133501.00/0
134 Journal of Speech and Hearing Research
found that stop-consonant identification was enhanced for
stimuli with dynamic onsets as compared to the static
spectral representation of Blumstein and Stevens (1980).
There has been little direct investigation of whether
psychoacoustic limitations of persons who are hearing
impaired affect their reliance on time-varying or static
spectra at stop-consonant onset. Only two studies have
used hearing-impaired subjects to examine the effect of
syllable duration on stop-consonant recognition (Dubno,
Dirks, & Schaefer, 1987; Van Tasell, Hagen, Koblas, &
Penner, 1982). Van Tasell et al. synthesized a five-formant voiced-stop place of articulation continuum and
then edited the stimuli to preserve only the first two
glottal periods following the release burst. Given substantial practice and a high presentation level (100 dB
SPL), normally hearing and mild to moderate hearingimpaired subjects showed at least 80% identification of
the exemplars on a/ba-do-oa/continuum.
Dubno et al. (1987) examined stop-consonant identification ability in persons who are hearing impaired using
syllable durations ranging from 10 ms to 100 ms. For/a/and
/u/ environments, subjects with fiat or gradually sloping
audiometrie configurations were able to achieve good place
categorization (>80%) after listening to only the first 20 ms
of the syllable. Subjects with steeply sloping losses demonstrafed significantly poorer performance even for long duration stimuli. Dubno et al. (1987) also examined syllables
in the/i/environment. Normally hearing subjects achieved
good identification with 20 to 40 ms durations. Hearingimpaired subjects had mean scores in the 40%-60% range
even at the longest syllable durations. The Van Tassell et al.
(1982) and Dubno et al. (1987) studies suggest that hearingimpaired subjects can, in some vowel environments at least,
use cues that are present within the first 20 to 40 ms of
stop-consonant syllables, just as do listeners with normal
hearing. The stimuli used in each of these studies were not
manipulated to remove formant-transition information, so it
is not known whether the hearing-impaired subjects relied
on the trajectory of the transition (even a substantially
h'uneated one) to make place decisions, or if static onset
information was sufficient for them to categorize the stimuli.
It is not known whether the frequency discrimination
ability of hearing-impaired subjects affects their ability to use
dynamic onset information. For steady-state pure-tone stimuli, the difference limen (DL) for frequency is elevated in
some hearing-impaired listeners (Butler & Albrite, 1957;
DiCarlo, 1962; Freyman & Nelson, 1987; Gengel, 1973; Hall
& Wood, 1984; Ross, Huntington, Newby, & Dixon, 1965;
Turner & Nelson, 1982). More importantly, this is also the
case for stimuli that have frequency transitions similar to
those encountered in speech (Arlinger, Jerlvall, Ahren, &
Holmgren, 1977; Collins, 1984; Danaher, Osberger, &
Pickett, 1973; Danaher & Pickett, 1975; Grant, 1987;
Martin, Pickett, & Colten, 1972; Pickett & Martony, 1970;
Tyler, Wood, & Fernandes, 1983; Van Tasell, 1980; Zurek
& Formby, 1981). The ability to discriminate frequency
changes may affect which aspects of stop-consonant onset
are used by the listener. If, for example, the frequency
change (AF) present in the formant transition does not
exceed the subject's DL, then stop-consonant syllables
32 133-142 March 1989
with dynamic onsets may provide no more information
than syllables without formant transitions. In this case,
formant transition information would simply be unavailable to the listener.
The experiments reported here sought to examine the
relationship between frequency discrimination ability and
perception of acoustic correlates of place of articulation.
Frequency discrimination data for normally hearing subjects suggest that the ability to detect a change in stimulus
frequency is poorest in stimuli that: (a) are short duration
(Chin-An & Chistovich, 1960; Freyman & Nelson, 1987;
Gengel, 1973; Hall & Wood, 1984; Henning, 1970; Moore,
1973; Tsumura, Sone, & Nimura, 1973; Turnbull, 1944);
(b) consist of many frequencies (Flanagan, 1955; Hoekstra
& Ritsma, 1977; Mermelstein, 1978; Van Tasell, 1980); and
(c) have spectra that change with time (Collins, 1984;
Mermelstein, 1978; Tyler et al., 1983; Van Tasell, 1980).
These characteristics are typical of stop consonants. The
stimuli used to investigate the relationship between stopconsonant perception and frequency discrimination, therefore, should mimic the characteristics of stop consonants as
much as possible. With this in mind, frequency discrimination for the second formant (F2) transition was examined
here in a syllable context.
The DL for F2 transitions has been measured in hearingimpaired subjects in several studies (Danaher et al., 1973;
Danaher & Pickett, 1975; Martin et al., 1972; Van Tasell,
1980). These data are of limited usefulness in explaining the
performance of subjects in the current speech-perception
experiments for two reasons. First, the DL for F2 has been
examined previously only in two frequency regions. Themfore the data could be used to predict performance for only a
few stimuli along a/b-d-g/continuum. Second, all data on
forrnant-transition DL in hearing-impaired subjects have
been obtained from listeners having moderate to profound
flat or slightly sloping hearing loss. This does not address the
very common configuration consisting of threshold elevation
in the high frequencies, with near normal sensitivity in the
lower frequency regions. The hearing-impaired subjects examined here had moderate high-frequency sensorineural
hearing loss, therefore the second formant of the stimuli used
was within the region of normal threshold. This stimulus
configuration is of interest because it has been shown that
abnormal frequency resolution can exist in the nomaal
threshold region of listeners with high frequency sensorineural hearing loss (Humes, 1983; Turner & Nelson, 1982).
EXPERIMENT
1
The purpose of Experiment 1 was to clarify which
acoustic properties of utterance-initial stop consonants
are important in identification of place of articulation by
hearing-impaired listeners, and by subjects in whom
hearing loss is simulated by the addition of masking
noise. This goal was accomplished by manipulating the
second formant transition of voiced stops so as to preserve
the onset spectrum, but remove the frequency transition,
as has been suggested by Stevens and Blumstein (Blumstein & Stevens, 1980; Stevens & Blumstein, 1978). In
OCHS ET AL.: Frequency Discrimination Ability
TABLE 1. Pure-tone air-conduction thresholds in dB HL for
hearing-impaired (H 1-H4) and masked-normal (M 1-M4) listeners.
kHz
H1
M1
H2
M2
H3
M3
H4
M4
0.5
1.0
2.0
3.0
4.0
6.0
5
5
10
55
75
65
15
8
16
46
75
73
15
15
10
55
70
75
10
10
20
51
75
65
10
10
5
50
60
60
15
6
2
49
57
52
5
10
5
65
65
60
10
9
2
60
68
62
addition, Experiment 1 examined stop-consonant perception for several syllable durations.
A group of masked-normal subjects was included to determine if any perceptual abnormalities observed in the hearing-impaired subjects could be explained purely on the basis
of elevated auditory thresholds and reduced dynamic range.
That is, if the threshold elevation and altered loudness
perception produced by the masking noise caused these
normally hearing listeners to perform like the hearing-impaired subjects, then one might assume that there were no
additional psychoaeoustic abnormalities, caused by the hearing loss, that were responsible for the errors in perception.
Method
Subjects and masking procedure. Four hearing-impaired
subjects were tested. In the test ear, all had pure-tone
air-conduction thresholds of15 dB HL or better (ANSI, 1969)
at octave frequencies from 250--2000 Hz, and poorer than 50
dB HL at 3000 Hz and above (see Table 1). In an attempt to
minimize intersubject variability, subjects were matched in
terms of pure-tone configuration and etiology. Subjects differed in their pure-tone thresholds by no more than 15 dB at
any frequency, and case history information suggested that
the hearing impairment was at least primarily noise-induced
in all cases. All subjects were between 30 and 55 years of age
and had no experience with amplification.
Eight subjects with normal hearing thresholds (ages 23
to 30 years) also participated. Four of these were presented with high-pass filtered noise throughout the experiments to simulate the audiometric configuration of
the hearing-impaired subjects.
Masking noise was provided by a Coulbourn noise
generator (Model $81-02) and was high-pass filtered by
two cascaded filters (Ithaco 4302) with a cut-offfrequency
of 3150 Hz and a combined rejection rate of 48 dB/octave.
Overall noise level for the masked-normal condition was
98 dB SPL for 3 of the 4 subjects (M1, M2, and M4), with
10
19
28
ii
44
i
I
3
g
II)
i
w
i
0
10
Formant
;I
20
n
l
m
~
/bi/
/di/
/gi/
F1
onset-target
276-307
276-307
276-307
F2
onset-target
1648--2070
1903-2070
2206-2070
m
l
30
Duration
m
i
40
i
~
i
50
(ms)
FIGURE 1. Schematic representation of the spectrographic analysis of/bi/. Formant structalre for the moving-F2 stimulus is
shown by the heavy lines. Straight-F2 stimuli were identical,
except that the second formant began at the appropriate frequency and was constant (dashed line). Both types of stimuli
were preceded by a burst. Numbers at the top of the figure
indicate the actual stimulus durations used in Experiment 1.
the level for Subject M3 being 88 dB SPL. Pure-tone
testing immediately before and after test sessions revealed that the masking noise was not intense enough to
have induced any temporary threshold shift. As is indicated in Table 1, the simulated losses did not differ from
the actual hearing losses by more than 10 dB at any
frequency. Typically, they differed by less than 5 dB.
Stimuli. Consonant-vowel syllables were generated using
a cascade/parallel synthesis program (Klatt, 1980) and an
LSI-11/23 microprocessor. The synthesized syllables were
output through a 12-bit D/A converter at a rate of 10 kHz, and
low-pass filtered at 4.0 kHz (rejection rate of 48 dB/octave).
A schematic representation of the spectrographic analysis
of/bi/is illustrated as an example in Figure 1. Fundamental
frequency rose from 105 Hz to 120 Hz over the first 30 ms of
voicing, and was then constant throughout the durations
relevant to these experiments. Onset and end-points of the
transitions, and all steady-state formant frequency values are
listed in Table 2. The five-formant stimuli had 40-ms formant
transitions present for the first and second forrnant (F1, F2)
only (shown by the solid, dark lines in Figure 1). F2 frequency did not vary linearly across the whole transition
duration. Frequency change was more gradual toward the
end of the transition. The short duration stimuli were edited
from the exemplars of a/bi-di-gi/continuum that had been
TABLE 2. Frequency parameters in Hz for stimuli in experiment 1.
Stimulus
135
F3
steady-state
F4
steady-state
F5
steady-state
2503
2708
2885
3300
3300
3300
3750
3750
3750
136 Journal of Speech and Hearing Research
shown to generate high phoneme categorization in this group
of hearing-impaired subjects. The nonlinear transitions were
maintained in this experiment to ensure that the stimuli
would evoke the best possible identification performance.
The first, fourth, and fifth formants were identical in all
stimuli. F3 had no transition, but rather was constant at the
onset frequency for each place of articulation. F3 was
controlled in this way to facilitate comparisons between
identification performance for these stimuli and formanttransition discrimination ability measured in Experiment 2.
Each stimulus was preceded by a noise burst that varied in
duration and spectrum as a function of place of articulation.
Each of the stimuli shown in Figure 1 was digitally
edited to provide six different stimulus conditions. Four
of these involved the burst plus 1, 2, 3, or 5 glottal pulses
corresponding, respectively, to the first 10, 19, 28, and 44
ms of voicing. Because F1 and F2 transitions were present, these stimuli (called moving-F2 stimuli) corresponded to the truncated stimuli with formant transitions
used by Blumstein and Stevens (1980). These stimuli
were replicated, except that no F2 transition was included, and were called straight-F2 stimuli. As is shown
in Figure 1 (see dashed line), straight-F2 stimuli were
exactly like moving-F2 stimuli except that the second
formant began at the appropriate onset frequency and was
then constant. Note that straight-F2 stimuli did have an
F1 transition, but it was the same throughout all stimulus
conditions, and identical to that in the moving-F2 stimuli.
Two additional conditions were included besides those
shown in Figure 1. In one, only the burst was retained
(called "burst" stimuli). In the other (called "burst +
vowel" stimuli), the burst was appended to a 10-ms segment of vowel (F1 = 307 Hz, F2 = 2070 Hz, F3 = 2980 Hz),
with a 10-ms portion of F1 transition included to give a
stop-like quality. These burst + vowel stimuli were more
stop-like than the burst-only stimuli, but did not supply F2
or F3 onset information specific to any place of articulation.
Procedures. Syllable identification was measured in a
single-interval, three-alternative forced-choice paradigm.
On each trial the subject heard a single syllable and was
asked to indicate whether the consonant was/b/,/d/, or/g/.
A most comfortable listening level (MCL) was established
for the longest duration stimulus, and all others were
presented at the same level. MCL varied from 70 dB to 85
dB SPL across subjects. The masked-normal subject listened at the same presentation levels as their matched
hearing-impaired subjects.
Stimuli were grouped so that only one duration occurred in a single test run. The order in which conditions
were presented was pseudorandom. A randomization was
discarded if it allowed presentation of three of the longest
duration stimuli consecutively, early in the testing.
Previously, the subjects tested in this experiment had
received approximately 15 hours of identification and
discrimination experience with full-duration, unmanipufated stimuli from/bo-do-ga/and/bi-di-gi/continua (Ochs,
1983). In the current experiment, all subjects received 10
trials per stimulus as practice. This practice differed from
that in similar study (Van Tasell et al., 1982) in that
subjects received no feedback in the identification tasks.
32 133-142
March 1989
Performance for the normally hearing subjects was calculated based on 10 trials per stimulus. Preliminary observation with hearing-impaired subjects revealed that they
experienced more difficulty in identifying place of articulation; therefore, the number of trials for hearing-impaired and masked-normal subjects was increased to 50
per stimulus to reduce variability in the data.
Results and Discussion
Figure 2 shows mean percent correct identification as a
function of stimulus duration for all moving-F2 stimuli, for
I
100
*6
80
~
so
i
I
i--
r
x
zx
40
0
¢J
D.
20
Ibil
0
I
I
8
B÷V
I
0
I
I
I
19 28
44
Formant Duration (:ms)
i
r
I
I
]
i
1
100
8O
©
L,.
O
o
60
~ 4o
~
2o
/di/
B÷V
0
I
J
19
28
/
44
Formant Duration (:ms)
I
]
o
~
[
I
I
I
100
80
==
3 60
o
4) 40
=u
0.
x
20
f
X NH
/gi/
~HI
©MN
0
B B+V
1
28
44
Forrnant Duration (ms)
FIGURE2. Mean identification performance for moving-F2 stimuli. Burst-only stimuli are denoted by B and burst + vowel
stimuli by B + V. Numerals along the abscissa refer to £ormant
duration. Three groups of subjects are shown: normally hearing
unmasked listeners (NH), normal hearers with masking noise
(MN), and hearing-impaired subjects (HI).
OCHS ET AL.: F r e q u e n c y D i s c r i m i n a t i o n A b i l i t y
the three groups of subjects. Means and standard deviations
for the three groups are presented in Table 3. Data were
arcsine transformed and then separate analyses of variance
(ANOVA) were performed on the data in each panel. For
purposes of analysis, the burst and burst + vowel stimuli
were considered as the first two steps on a continuum of
increasing stimulus duration. Performance improved significantly with duration for/bi/,/dY, and/9i/, [/bi/: F(5,45) =
15.94, p < .0001;/di/: F(5,45) = 18.79, p < .0001;/9i/: F(5,45)
= 9.99, p < .0001]. A Duncan post hoc analysis of the
duration effect showed that for/bi/, asymptotic performance
(44 ms) was significantly (p < .05) greater than that for all
other durations except 28 ms. For/di/, performance at 44 ms
was significantly (p < .05) greater than performance for all
but the 19 ms and 28 ms durations. For/gi/, performance at
44 ms was significantly (p < .01) greater than that for the
burst (B) stimuli.
A between-gr6ups analysis of the data in Figure 2
revealed marginally significant differences for the moving-F2 /bi/ stimulus only, [F(2, 9) = 3.81, p < .06].
Performance for normally bearing u n m a s k e d listeners
was most accurate, and performance for masked-normal
subjects least accurate. No significant difference b e t w e e n
groups was observed for /di/ and /gi/. No significant
interactions were o b s e r v e d for the data presented in
Figure 2.
TABLE 3. Mean percent correct consonant identification for moving-F2 and straight-F2 stimuli.
Standard deviations are in parentheses.
Condition
/bi/
normal-hearing
moving-F2
B
B+ V
10 ms
19 ms
28 ms
44 ms
95.0
(5.8)
72.5
(22.2)
100.0
(0.0)
97.5
(5.0)
70.0
(46.9)
85.0
(19.2)
97.5
(5.0)
100.0
(0.0)
100.0
(0.0)
100.0
(0.0)
36.9
(45.1)
42.5
(*28.4)
59.4
(40.0)
64.4
(41.9)
66.9
(27.3)
71.3
(37.9)
76.3
(39.3)
78.8
(34.7)
97.5
(3.5)
71.3
(47.5)
46.9
(36.1)
40.0
(17.2)
75.0
(22.5)
63.1
(20.7)
84.4
88.8
(21.0)
84.4
(16.6)
98.8
straight-F2
masked-normal
moving-F2
straight-F2
hearing-impaired
moving-F2
straight-F2
/di/
normal-hearing
moving-F2
77.5
47.5
90.0
(15.0)
(15.0)
(20.0)
85.0
straight-F2
masked-normal
moving-F2
69.4
(31.9)
60.9
(24.8)
straight-F2
hearing-impaired
moving-F2
/9i/
normal-hearing
moving-F2
straight-F2
(20.7)
90.0
95.0
(14.1)
92.5
(10.0)
100.0
(12.9)
i00.0
(0.0)
90.0
(14.1)
(9.6)
(0.0)
70.6
(11.4)
70.0
96.9
(4.7)
98.8
97.5
(3.5)
93.1
97.5
(3.5)
96.9
(5.0)
(1.4)
(5.2)
(3.8)
59.4
(15.2)
44.4
(19.8)
71.9
(18.5)
83.1
(11.1)
82.5
(23.0)
88.1
(14.8)
92.5
(4.1)
86.3
(11.3)
27.5
(31.0)
70.0
(14.1)
82.5
(35.0)
87.5
(25.0)
95.0
(10.0)
100.0
(0.0)
82.5
(35.0)
80.0
(40.0)
77.5
(45.0)
87.5
(18.9)
69.4
(25.9)
71.9
(24.5)
91.9
99.4
99.4
100.0
(9.9)
90.6
(18.8)
(1.3)
97.5
(5.0)
(1.3)
100.0
(0.0)
100.0
81.3
(31.0)
83.1
(20.1)
98.1
99.4
100.0
(i.3)
98.8
(1.4)
(1.3)
98.8
(2.5)
100.0
(0.0)
straight-F2
hearing-impaired
moving-F2
(2.5)
67.5
47.5
(22.2)
straight-F2
masked-normal
moving-F2
(8.0)
77.5
(18.6)
55.0
(33.8)
straight-F2
68.8
(37.6)
73.8
(40.3)
137
(0.0)
(0.0)
(0.0)
138 Journal of Speech and Hearing Research
I
I
I
i
3o
._~
20
a.
lO
I
c
•~
o
-10
I
I
I
10
19
28
i
44
Formant Duration (ms)
I
I
I
I
30
/di/
E
g
20
¢1
10
I
c
32 133-142
In summary, Experiment 1 revealed that as stimulus
duration increased, all subject groups showed improved
identification performance for moving-F2 stimuli. Identification of these stimuli was basically the same for all groups
for the/di/and/9i/stimuli. Note, however, that the mean
data in Figure 2 reveal slightly poorer performance for the
normally hearing subjects for the longer durations of/gi/.
This was due to the very poor identification accuracy for a
single normally hearing subject on these conditions.
A significant group effect was observed for/bi/stimuli.
Normally hearing subjects had the best performance,
whereas masked-normal listeners exhibited the poorest
performance. In addition, Experiment 1 revealed that
overall, the groups do not differ in their performance for
moving-F2 versus straight-F2 stimuli. Only at the longest
duration o f / b i / w a s performance for moving-F2 stimuli
better than that for straight-F2 stimuli for the hearingimpaired and masked-normal listeners. Experiment 2 was
undertaken to explore perceptual limitations that may
have produced the pattern of results observed here.
0
O
$
EXPERIMENT
I
I
I
I
10
19
28
44
I
._~
30
I
i
l
×NH
/g~/
ZXHI
20
Method
10
f-
'.~
O
Experiment 2 examined frequency-discrimination functions for the second formant (F2). It was hypothesized that
frequency discrimination ability would limit the extent to
which subjects could use second formant transition information to enhance consonant identification.
OMN
E
I
2
-10
Formant Duration (ms)
"N
March 1989
0
-10
I
10
/
19
I
I
28
44
Formant Duration (ms)
FIGURE 3. Mean difference in identification performance between stimuli with moving-F2 versus straight-F2 [P(c) = percent
correct] for three groups of subjects: normally hearing umnasked
listeners (NH), normal hearers with masking noise (MN), and
hearing-impaired subjects (HI).
Figure 3 displays the performance difference observed
for stimuli with moving-F2 versus straight-F2, for all
groups of subjects (see Table 3 also). The mean percent
correct score for stimuli with straight-F2 was subtracted
from that for moving-F2 stimuli and is presented on the
ordinate. A positive number, therefore, reveals better
performance for moving-F2 versus straight-F2 stimuli. An
ANOVA was performed on the arcsine transformed data
for each place of articulation to look for between-group
differences and differences as a function of duration. A
significant main effect for duration was obtained for/bi/
only, [F(3,27) = 4.98, p < .007], due to a beneficial effect
of F2 movement for masked-normal and hearing-impaired subjects at the 44-ms duration.
Subjects. Subjects were the same as those used in
Experiment 1.
Stimuli. To compare the identification performance
measured in Experiment i with the data generated in this
experiment, difference limen for second formant frequency was examined for stimuli having onset frequencies and formant structures similar to/bi/,/di/, and/gi/. In
addition, DL was assessed at stimulus durations roughly
corresponding to those used in Experiment 1 (10, 20, 30,
50 ms). Stimuli were generated with the same software
and filtering arrangement described previously.
A schematic example of the spectrogram of 50-ms
standard and variable stimuli (comparable t o / b i / i n Experiment 1) is shown in Figure 4. Stimuli had 5 formants,
with F4 and F5 the same throughout the experiment. F3
had no frequency transition, but its steady-state frequency was appropriate for the syllable under study. In
all standard and variable stimuli, F1 varied from 276 Hz
to 307 Hz over the duration of the stimulus.
F2 transition DL was measured in an adaptive paradigm
(Levitt, 1971). Each syllable and duration condition, therefore, required a series of stimuli with varying degrees of F2
transition (AF). The onset frequency for F2 was appropriate
for each consonant involved, but in each series, the AF for
F2 ranged from 3 Hz to 400 Hz in 20 logarithmic steps. The
step size between AFs thus ranged from 2 to 3 Hz at a small
AF, to a maximum of 91 Hz at a AF of 400 Hz. For stimuli
OCHS ET AL.: Frequency Discrimination Ability
g
u
i
139
|
variable
standard
Z
(J
C
G)
O"
I
i
i
10
20
30
i
40
i
50
I
I
i
i
0
10
20
30
i
40
I
50
Duration ( m s )
FIGURE 4. Schematic representation of the spectrographic analysis of a standard and a variable
stimulus used for frequency DL measures for/bi/. The standard stimulus has five formants, with
a 31 Hz formant transition in F1. All other formants have no transition. The variable stimulus is
identical except that the endpoint of the second formant is varied adaptively during testing
(dashed lines).
comparable to /bi/ and /di/, the most extreme F2 target
frequency was 400 Hz higher than onset frequency,
whereas for/gi/it was 400 Hz lower. These F2 frequency
excursions extended over four different stimulus durations
to allow an examination of this variable. Unavoidably,
therefore, rate of frequency change was not constant across
duration conditions. The maximum difference between the
standard stimulus and the variable one occurred at the end
of the stimulus because F2 began at the appropriate frequency for that phoneme. This approach was chosen because it is the most accurate simulation of the acoustic
properties characterizing the moving-F2 and straight-F2
stimuli in Experiment 1.
Procedures. Stimuli were presented in a four-interval
oddity paradigm. This involved presentation of two pairs
of stimuli, one of which contained an "odd" stimulus.
The odd stimulus was a variable stimulus with some
degree olAF present. The other three intervals contained
a straight-F2 stimulus with the appropriate F2 onset
frequency (see Figure 4). The odd stimulus could occur
in any of the four intervals. F2 D L was determined
adaptively converging on the 71% point of the psychometric function (Levitt, 1971).
Stimuli were presented at MCL, with presentation
levels ranging from 70 to 80 dB SPL across subjects. The
longest stimuli were used to arrive at MCL, and all other
stimuli were presented at the same equipment settings.
This resulted in an average absolute threshold difference
of 4 dB between the shortest and longest stimuli.
Order of presentation of stimulus conditions was random. Data from the first stimulus condition tested were
regarded as practice, and the condition was repeated at a
later time.
Results and discussion
Mean F2 transition DLs for the three groups of subjects
are shown in Figure 5 (see Table 4 for standard devia-
tions). The dark diagonal line overlaying the data in each
panel is a schematic representation of the frequency
change covered by the F2 transition in each of the
moving-F2 stimuli examined in Experiment 1. Recall that
F2 transitions were not linear across the whole transition
duration in Experiment 1. When data points occur below
the diagonal line, this indicates that frequency discrimination was acute enough to detect the presence of the
transition. If the data points occur above the diagonal
line, then D L for F2 transitions exceeded the ~F present
in the stimulus at that duration.
Figure 5 reveals that the DL for F2 transition does not
differ between the three groups of subjects tested. For
/bi/, the ~F present in the F2 transition for stimuli in
Experiment 1 is clearly greater than the mean D L for all
subject groups. That is, the F2 transition information in
the/bi/stimuli from the previous experiment should have
been easily available to the listeners. F o r / d i / a n d / g i / ,
however, the F2 frequency change present in the Experiment 1 stimuli was not substantially greater than that
required for discrimination threshold. In these cases, the
DL data suggest that frequency-discrimination ability
may have limited the perception of moving-F2 transitions.
GENERAL
DISCUSSION
Experiment 1 revealed that for/di/and/gi/stimuli with
a moving-F2, these normally hearing, masked-normal,
and high-frequency hearing-impaired subjects performed
similarly. Performance improved with duration and was
close to 100% at the longest durations. Results for the
hearing-impaired subjects are better than those observed
previously for a similar hearing loss configuration (Dubno
et al., 1987).
For the /bi/ stimuli, identification performance improved with duration, but hearing-impaired subjects
showed slightly poorer recognition ability than the nor-
140 Journal o f Speech and Hearing Research
32
/bi/
Stimulus duration
300
-r"
a
Condition
200
100
0
0,
1;
t
20
'
30
40
'
50
Duration (ms)
1
i
I
I
~
I
I
4oo:
/di/
/bi/
normal-hearing
masked-normal
hearing-impaired
/dU
normal-hearing
masked-normal
hearing-impaired
/oil
normal-hearing
masked-normal
hearing-impaired
10 ms
20 ms
30 ms
50 ms
83 (56)
104 (37)
98 (35)
52 (12)
60 (24)
79 (27)
44 (7)
85 (44)
68 (21)
41 (18)
67 (36)
73 (50)
94 (35)
76 (47)
91 (52)
117 (52)
112 (43)
94 (37)
84 (49)
85 (17)
71 (31)
72 (19)
79 (42)
88 (31)
a
a
a
77 (35)
107 (34)
79 (26)
93 (57)
113 (42)
99 (19)
118 (20)
156 (49)
131 (34)
aData are incomplete for this condition.
~vv
•
J
c~
March 1989
TABLE 4. Mean frequency difference limen for/bi/,/di/, and/gi/.
Standard deviations are in parentheses.
400
,.J
133-142
200
~oo
t
0
1to
I
I
I
1
20
30
40
50
Duration ( m s )
f
400
/gi/
/gl.
x NH
300
© MN
z~HI
200
lOO
o
' ' d10o ; ' '
0
3
40
would have expected more errors for /gi/, because its
second formant has a higher frequency. The data of
Dubno et al. (1987) suggest another explanation. These
authors report that both /b/ and /d/ are more poorly
identified than /g/ in subjects with sloping high-frequency hearing losses, as compared to subjects with fiat
losses. They suggest that a sloping hearing loss may cause
a more radical distortion of the onset spectrum o f / b / a n d
/d/than it does with the more compact onset spectrum of
50
Duration ( m s )
FIGURE 5. Mean second formant frequency difference limen for
three groups of subjects. The heavy diagonal line schematically
represents the extent of frequency change encompassed by the
formant transition in the moving-F2 stinmli examined in Experiment 1. Only the transition extent (AF) is shown. The transition
was from low to high frequencies for/bi/and/di/, and from high
to low frequencies for/gi/. NH = normally hearing unmasked
subjects, MN = normally hearing masked subjects, HI = hearing-impaired subjects. In the bottom panel (/9i/), no data were
available in the 10-ms condition, due to a synthesis error.
mally hearing subjects, at least at the shorter durations.
Masked-normal listeners had even more difficulty. When
errors were m a d e , / b i / w a s confused most often with/di/.
This confusion b e t w e e n / b i / a n d / d i / i s somewhat surprising because hearing-impaired and masked-normal subjeets had normal pure-tone thresholds through 2000 Hz.
The second formants of /bi/ and /di/, therefore, were
within the area of normal thresholds. If the errors were
the result of inadequate stimulation level, then one
Another possible explanation for such findings is that
psychoacoustic limitations prevented accurate consonant
recognition. The difference limen for F2 transition was
examined in Experiment 2 to investigate its role in stopconsonant identification. Earlier work suggested that some
subjects with high-frequency sensorineural hearing loss
have a poorer than normal difference limen for frequency in
areas of normal pure-tone thresholds (Turner & Nelson,
1982). In the present study, frequency DLs were equivalent
in hearing-impaired, masked-normal, and normally hearing
subjects, therefore it would seem that frequency discrimination ability cannot explain group differences in consonant
identification performance.
Experiment 1 revealed that for the subjects examined
here, the presence of a moving-F2 did not have a large
effect on consonant identification. For short-duration/bi/
stimuli, and all/di/and/gi/stimuli, a moving-F2 produced
no better performance than did a straight-F2. Only in the
case of a 44-ms /bi/ stimulus did hearing-impaired and
masked-normal subjects identify moving-F2 stimuli better than those with a straight-F2. The difference limen
data reveal that in the case of/di/and/gi/, there was not
enough frequency change in the F2 transition for it to be
discriminated reliably from the straight-F2 stimulus. The
moving-F2 and straight-F2 stimuli were, in effect, perceptually equivalent, therefore it is not surprising that no
performance difference was observed. This was not the
case with /bi/, however, because the frequency change
encompassed by F2 was greater than the subjects' frequency DL, even at short durations. This meant that the
OCHS ET AL.: Frequency Discrimination Ability
F2 transition was available as a cue for 19-ms, 28-ms, and
44-ms/bi/stimuli. It is interesting to note, however, that
masked-normal and hearing-impaired subjects had enhanced identification for moving-F2 stimuli in only the
44-ms condition. At shorter durations, identification was
not perfect and both groups could have benefitted from
additional cues.
One explanation for the uniform performance observed
for moving-F2 and straight-F2 stimuli is that these subjects
did not use the formant transition to identify the consonant.
This would be consistent with the view that the information
necessary to determine place of articulation for stop consonants is contained in the onset spectrum of the syllable
(Blumstein & Stevens, 1979; Blumstein & Stevens, 1980).
These authors have suggested, however, that listeners can
use moving-F2 information to maximize performance in
difficult listening situations. Given the less than perfect
identification scores observed here, one would expect that
subjects would have used the formant transition to enhance
recognition. It has recently been noted that when twoformant stimuli are presented to normally hearing subjects,
identification of place of articulation for stops is high only
when transition movement encompasses more than 200 Hz
(Ohde, 1988). Moreover, this study revealed that even when
the transition was substantial, the F2 transition was not a
sufficient condition for labial and alveolar identification.
These studies suggest that the F2 transition is not necessarily helpful in determining place of articulation.
An alternative explanation is that frequency discrimination ability was overestimated in this study, and that
actually the formant transitions in identification stimuli
were not detected. Although the stimuli used to determine DL were very similar to the identification stimuli,
there were some differences. For example, the stimuli
used for identification were preceded by a burst, whereas
the DL stimuli were not. A more recent investigation
suggests that F2 DL is substantially poorer in some
subjects when the second formant transition is preceded
and followed by acoustic energy (Ochs, 1987). In addition, it has been shown that frequency discrimination
ability is poorer when the standard and variable stimuli
both change in frequency (Collins, 1984; Tyler et al.,
1983). In this study, the standard stimuli had only a 31 Hz
frequency transition in F1, and all other formants were
stationary. This may have allowed better frequency discrimination ability than was possible with the more
complex speech stimuli used in Expriment 1. These data
point out the importance of using appropriate, complex
stimuli for assessment ofpsychoacoustic ability when one
attempts to explain speech perceptual errors in hearingimpaired subjects.
ACKNOWLEDGMENTS
This research represents a portion of the doctoral dissertation
of the first author, which was completed at Vanderbilt University. Some of the results of these experiments were presented
before the Acoustical Society of America at the 105th meeting in
Cincinnati, Ohio (1983). The authors wish to thank the subjects
for their generous participation. We wish to thank Mary Joe
141
Osberger, Mary Florentine, and an anonymous reviewer for their
helpful comments on an earlier draft of this manuscript. The first
author's doctoral level training was supported by Maternal &
Child Health Training grant #00028.
REFERENCES
American National Standards specifications for audiometers
(ANSI $3.6-1969). New York: ANSI.
ARLINGER, S. D., JERLVALL,L. B., AHREN, T., & HOLMGREN,
E. C. (1977). Discrimination of frequency ramps in subjects
with cochlear hearing loss. Acta Oto-laryngologica, 83,
310-316.
BILGER, R. C., & WANG,M. D. (1976). Consonant confusions in
patients with sensorineural hearing loss. Journal of Speech
and Hearing Research, 19, 718-748.
BLUMSTEIN,S. E., & STEVENS,K. N. (1979). Acoustic invariance
in speech production: Evidence from measurements of the
spectra characteristics of stop consonants. Journal of the
Acoustical Society of America, 66, 1001-1017.
BLUMSTEIN, S. E., & STEVENS, K. N. (1980). Perceptual invariance and onset spectral for stop consonants in different vowel
environments. Journal of the Acoustical Society of America,
67, 648-662.
BUTLER,R. A., & ALBRITE,J. P. (1957). The pitch discrimination
function of the pathological ear. Archives of Otolaryngology,
63, 411-418.
Buus, S., MILLER,J. L., & SCHARF, B. (1983). Masking of the
/da/-/ga/ distinction in cochlear impairment. Journal of the
Acoustical Society of America, 74(Suppl. 1), 111.
BYERS, V. (1973). Initial consonant intelligibility by hearingimpaired children. Journal of Speech and Hearing Research,
16, 48-55.
CHIN-AN,L., & CHISTOVlCH,L. A. (1960). Frequency-difference
limens as a function of tonal duration. Soviet Physical Acoustics, 6, 75-80.
COLLINS, M. (1984). Tone-glide discrimination: Normal and
hearing-impaired listeners. Journal of Speech and Hearing
Research, 27, 403--412.
DANAHER, E. M., OSBERGER, M.J., & PICKETT, J. M. (1973).
Discrimination of formant frequency transitions in synthetic
vowels. Journal of Speech and Hearing Research, 16, 439-451.
DANAHER,E. M., & PICKETT,J. M. (1975). Some masking effects
produced by low frequency vowel formants in persons with
sensorineural hearing loss. Journal of Speech and Hearing
Research, 18, 261-271.
DANHAUER,J., & LAWARRE,R. M. (1979). Dissimilarity ratings
of English consonants by normally hearing and hearing-impaired individuals. Journal of Speech and Hearing Research,
22, 236-246.
DICARLO,L. M. (1962). Some relationships between frequency
discrimination and speech reception performance. Journal of
Auditory Research, 2, 37-49.
DORMAN, M. F., STUDDERT-KENNEDY,M., & RAPHAEL,L.J.
(1977). Stop-consonant recognition: Release bursts and formant transitions as functionally equivalent, context-dependent
cues. Perception and Psychophysics, 22, 109-122.
DRESCHLER, W.A., & PLOMP, R. (1985). Relations between
psychophysical data and speech perception for hearing-impaired subjects II. Journal of the Acoustical Society of America, 78, 1261-1270.
DUBNO,J. R., DIRKS,D. D., & SCHAEFER,A. B. (1987). Effects of
hearing loss on utilization of short-duration spectral cues in
stop consonant recognition. Journal of the Acoustical Society
of America, 81, 1940-1947.
FLANAGAN,J. L. (1955). A difference limen for vowel formant
frequency. Journal of the Acoustical Society of America, 27,
613-617.
FREYMAN, R. L., & NELSON, D. A. (1987). Frequency discrimination of short- versus long-duration tones by normal and
hearing-impaired listeners. Journal of Speech and Hearing
142 Journal of Speech and Hearing Research
Research, 30, 28-36.
GENGEL,R. W. (1973). Temporal effects in frequency discrimination by hearing-impaired listeners. Journal of the Acoustical
Society of America, 54, 11-15.
GORDON-SALANT, S.M., & WlGHTMAN, F.L. (1983). Speech
competition effects on synthetic stop-vowel perception by
normal and hearing-impaired listeners. Journal of the Acoustical Society of America, 73, 1756-1765.
GRANT, K. W. (1987). Frequency modulation detection by normally hearing and profoundly hearing-impaired listeners.
Journal of Speech and Hearing Research, 30, 558--563.
HALL, J. W., & WOOD, E. J. (1984). Stimulus duration and frequency discrimination for normal-hearing and hearing-impaired subjects. Journal of Speech and Hearing Research, 27,
252-256.
HANNLEY, M., & DORMAN, M. F. (1983). Susceptibility to intraspeech spread of masking in listeners with sensorineural
hearing loss. Journal of the Acoustical Society of America, 74~
40-51.
HENNING, G. B. (1970). A comparison of the effects of signal
duration on frequency and amplitude discrimination. In R.
Plomp & G. F. Smoorenburg (Eds.), Frequency analysis and
periodicity detection in hearing (pp. 350-361). Leiden: A. W.
Sijthofi:
HOEKSTRA, A., & RITSMA, R. J. (1977). Perceptive hearing loss
and frequency selectivity. In E. F. Evans & J. P. Wilson (Eds.),
Psychophysics and physiology of hearing (pp. 263-271). London: Academic Press.
HUMES, L. E. (1983). Mid-frequency dysfunction in listeners
having high frequency sensorineural hearing loss. Journal of
Speech and Hearing Research, 26, 425-435.
KEWLEY-PORT,D. (1983). Time-varying features as correlates of
place of articulation in stop consonants. Journal of the Acoustical Society of America, 73, 322-335.
KEWLEY-PORT, D., PISONI, D. B., & STUDDERT-KENNEDY,M.
(1983). Perception of static and dynamic acoustic cues to place
of articulation in initial stop consonants. Journal of the Acoustical Society of America, 73, 1779-1793.
KLATT, D. H. (1980). Software for a cascade/parallel formant
synthesizer. Journal of the Acoustical Society of America, 67,
971-990.
LAHIRI,A., GEWITH, L., & BLUMSTEIN,S. E. (1984). A reconsideration of acoustic invariance for place of articulation in diffuse
stop consonants: Evidence from a cross-language study. Journal of the Acoustical Society of America, 76, 391-404.
LEVITT, H. (1971). Transformed up-down methods in psychoacoustics. Journal of the Acoustical Society of America, 49,
467-477.
MARTIN, E. S., PICKETT, J. M., & COLTEN, S. (1972). Discrimination of vowel formant transitions by listeners with severe
sensorineural hearing loss. In G. Fant (Ed.), Speech communication ability and profound deafness (pp. 81-98). Washington DC: A. G. Bell Association.
MERMELSTEIN,P. (1978). Difference limens for formant frequencies of steady-state and consonant-bound vowels. Journal of
the Acoustical Society of America, 63, 572-580.
MOORE, B. C. J. (1973). Frequency difference limens for short
duration tones. Journal of the Acoustical Society of America,
54, 610-619.
OCHS, M.T. (1983). The perception of stop consonants by
listeners having high frequency sensorineural hearing loss.
Unpublished doctoral dissertation, Vanderbilt University,
Nashville, Tennessee.
OCHS, M. T. (1987). Frequency discrimination and consonant
recognition in the hearing impaired. Journal of the Acoustical
Society of America, 82, (Supp. 1), 4.
OHDE, R. N. (1988). Revisiting stop-consonant perception for
two-formant stimuli. Journal of the Acoustical Society of
America, 84, 1551-1555.
OWENS, E., BENEDICT, M., & SCHUBERT, E. D. (1972). Conso-
32
133-142
March 1989
nant phonemic errors associated with pure-tone configuration
and certain kinds of hearing impairment. Journal of Speech
and Hearing Research, 15, 308--322.
OWENS, E., & SCHUBERT, E. D. (1968). The development of
consonant items for speech discrimination testing. Journal of
Speech and Hearing Research, 11,656-667.
PICKETT, J. M., & MARTONY, J. (1970). Low-frequency vowel
formant discrimination in hearing-impaired listeners. Journal
of Speech and Hearing Research, 13, 347-359.
REED, C. (1975). Identification and discrimination of vowelconsonant syllables in listeners with sensorineural hearing
loss. Journal of Speech and Hearing Research, 18, 773-794.
Ross, M., HUNTINGTON,D. A., NEWBY, H. A., & DIXON, R. F.
(1965). Speech discrimination of hearing-impaired individuals
in noise. Journal of Auditory Research, 5, 47-72.
SHER, A. E., & OWENS, E. (1974). Consonant confusions associated with hearing loss above 2000 Hz. Journal of Speech and
Hearing Research, 17, 669--681.
STEVENS, K. N., & BLUMSTEIN,S. E. (1978). Invariant cues for
place of articulation in stop consonants. Journal of the Acoustical Society of America, 64, 1358-1368.
TSUMURA,T., SONE, T., & NIMURA,T. (1973). Auditory detection
of frequency transition. Journal of the Acoustical Society of
America, 53, 17-25.
TURNBULL, W. W. (1944). Pitch discrimination as a function of
tonal duration. Journal of Experimental Psychology, 34,
302--316.
TURNER, C. W., & NELSON, D. A. (1982). Frequency discrimination in regions of normal and impaired sensitivity. Journal of
Speech and Hearing Research, 25, 34--41.
TURNER, C.W., & VAN TASELL, D.J. (1984). Sensorineural
hearing loss and the discrimination of vowel-like stimuli.
Journal of the Acoustical Society of America, 75, 562--565.
TYLER, R. S., SUMMERFIELD,Q., WOOD, E. J., & FERNANDES,M.
(1982). Psychoacoustic and phonetic temporal processing in
normal and hearing-impaired listeners. Journal of the Acoustical Society of America, 72, 740-752.
TYLER, R. S., WOOD, E. J., & FERNANDES,M. (1983). Frequency
resolution and discrimination of constant and dynamic tones in
normal and hearing-impaired listeners. Journal of the Acoustical Society of America, 74, 1190-1199.
VAN TASELL, D. J. (1980). Perception of second formant transitions by hearing-impaired persons. Ear and Hearing, 1,
130-136.
VAN TASELL,D.J., HAGEN,L. T., I(OBLAS,L. L., & PENNER,
S. G. (1982). Perception of short-term spectral cues for stop
consonant place by normal and hearing-impaired subjects.
Journal of the Acoustical Society of America, 72, 1771-1780.
WALDEN, B. E., & MONTGOMERY,A. A. (1975). Dimensions of
consonant perception in normal and hearing-impaired listeners. Journal of Speech and Hearing Research, I8, 444--455.
WALLEY, A. C., & CArmELL, T. D. (1983). Onset spectra and
formant transitions in the adult's and child's perception of
place of articulation in stop consonants. Journal of the Acoustical Society of America, 73, 1011-1022.
WANG, M. D., REED, C. M., & BILGER, R. C. (1978). A comparison of the effects of filtering and sensorineural hearing loss on
patterns of consonant confusions. Journal of Speech and Hearing Research, 21, 5--36.
ZUREK, P. M., & FoPavn3y, C. (1981). Frequency-discrimination
abi!ity of hearing-impaired listeners. Journal of Speech and
Hearing Research, 24, 108-112.
Received August 7, 1987
Accepted May 23, 1988
Requests for reprints should be sent to Marleen T. Ochs,
Division of Hearing and Speech Sciences, Vanderbilt University
School of Medicine, Nashville, TN 37232.