Auditory filter shapes in normal-hearing, noise-masked normal, and elderly listeners a)

Mitchell S. Sommers and Larry E. Humes Speech Research Laboratory and Departments of Psychology and Speech and Hearing Sciences, Indiana Unioersity, Bloomington, Indiana 47405

(Received 23 July 1992; accepted for publication 4 January 1993)

To dissociate the effects of age and hearing impairment on changes in frequency selectivity, auditory filter shapes were measured at 2 kHz in four groups of subjects: (1) normal-hearing young subjects; (2} normal-hearing elderly subjects; (3) elderly hearing-impaired listeners; and (4) young normal-hearing listeners with simulated hearing losses. Filter shapes were derived using a modified version of the notched-noise procedure [Glasberg and Moore, Hear, Res. 47, 103-138 (1990)]. Equivalent rectangular bandwidths (ERBs) of auditory filters were not significantly different in young and elderly subjects with normal 2-kHz hearing. Furthermore, filter widths for young subjects with 20- and 40-dB simulated hearing losses overlapped with those obtained from elderly subjects with corresponding degrees of actual hearing loss. One measure that did show significant differences between actual and simulated hearing losses was the degree of filter asymmetry; auditory filters in hearing-impaired listeners were more asymmetrical than those obtained from noise-masked normal-hearing subjects. The dynamic range of auditory filters, however, was comparable for hearing-impaired and noise-masked listeners. Lastly, post-filter detection efficiency was also similar for young and elderly subjects with equivalent hearing levels. These findings suggest that the reduced frequency selectivity often reported for older listeners can be attributed, primarily, to hearing loss rather than increased age. Implications of the results for speech perception in the elderly and models of hearing impairment are discussed.

PACS numbers: 43.66.Dc, 43.66.Sr [WDW]

INTRODUCTION

The peripheral auditory system has traditionally been modeled (Fletcher, 1940) as containing a series of over- lapping bandpass filters, the auditory filters, with approx- imately linear response characteristics. The shape of these filters can be considered a transfer or weighting function imposed upon the amplitude spectrum of a sound. Consid- erable evidence now exists that the selectivity of auditory filters can be altered by a number of diverse factors, includ- ing cochlear impairment (Zwicker and Schorn, 1978; Flo- rentine eta!., 1980; Glasberg and Moore, 1986; Tyler, 1986), anoxia (Robertson and Manley, 1974; Evans, 1975), and noise (Tyler et al., 1980, 1982). Additionally, a number of investigations (Patterson et al., 1982; Lutman and Clark, 1986) have demonstrated that auditory filters tend to broaden as a function of age. This finding of age- related reductions in frequency selectivity has been used to account for the decreased intelligibility of speech in noise often reported for older listeners (Kalikow eta!., 1977; Plomp and Mimpen, 1979; Hannley and Dorman, 1983; Dubno et al., 1984) compared to younger subjects with similar absolute thresholds. Specifically, the increased filter widths for older individuals are hypothesized to decrease

a•Portions of these data were presented at the 123rd meeting of the Acous- tical Society of America [J. Acoust. Soc. Am. 91, 2458(A) (1992)].

the signal-to-noise ratio at the output of auditory filters and therefore make identification of speech signals more diffi- cult.

The results of studies investigating age-related changes in auditory-filter shapes, however, are often difficult to in- terpret because decreased frequency resolution has been associated with both hearing impairment [see Tyler (1986) or Humes et al. (1988) for a review of frequency resolu- tion in hearing-impaired listeners] and increased age (Patterson etal., 1982). Therefore, given the well- documented changes in hearing thresholds due to presby- cusis (Hinchcliffe, 1962; Corso, 1971), most investigations of the effects of age on auditory-filter shapes have been confounded by age-related changes in absolute sensitivity. Patterson et al. (1982), for instance, used a notched-noise masker (Patterson, 1976) to determine the shape of audi- tory filters in young and elderly subjects at 0.5, 2.0, and 4.0 kHz. They demonstrated a significant positive correlation between age and the width of auditory filters at each of these frequencies and suggested that frequency selectivity declines as a function of age. However, since all of the older listeners had some degree of hearing impairment, it was not possible to assess the independent contributions of age and presbycusis to changes in auditory-filter shapes. Lutman et al. ( 1991 ) obtained two measures of frequency selectivity, psychophysical tuning curves (PTCs) and notched-noise thresholds, from subjects over age 50 with hearing levels between 0 and 80 dB HL. Although PTCs showed a greater upward spread of masking for older lis-

2903 d. Acoust. Soc. Am. 93 (5), May 1993 0001-4966/93/052903-12506.00 ¸ 1993 Acoustical Society of America 2903

teners, the effects of age per se could not be determined because of associated age-related differences in hearing sen- sitivity. Results from the notched-noise experiment dem- onstrated that, while both hearing impairment and age contributed to broader auditory filters, the two factors ac- counted for 25% and 4% of the variance, respectively. This finding suggests that hearing loss, rather than age, is the principal factor accounting for the reduced frequency resolution that has been reported for older listeners.

This conclusion is supported by results from a number of recent experiments (Peters and Moore, 1992a,b; Peters et al., 1992) designed to dissociate the effects of hearing loss and age on frequency selectivity. Peters and Moore (1992a) measured auditory-filter shapes in young and eld- erly subjects at frequencies between 100 and 800 Hz where both groups had essentially normal hearing. With the ex- ception of the lowest frequency (100 Hz), they found no significant difference in equivalent rectangular bandwidths (ERBs) as a function of age. They did, however, obtain differences in the efficiency with which the signals were detected after filtering; older listeners displayed reduced detection efficiency relative to younger subjects. In a see- ond study, Peters and Moore (1992b) reported that ERBs were not significantly different in young and elderly hearing-impaired individuals with similar degrees of hear- ing loss. In contrast to the earlier investigation, however, they did not find significant differences in detection effi- ciency as a function of age.

It is important to note that while recent studies of age-related changes in frequency selectivity (Peters and Moore 1992a,b) implicate absolute sensitivity as the prin- cipal factor contributing to broader auditory-filter shapes in elderly listeners, they have not been designed to examine whether hearing loss or specific cochlear and retrocochlear pathologies account for the changes. For example, the re- sults of Peters and Moore (1992b), demonstrating that younger and older subjects with comparable hearing losses exhibit similar changes in auditory filters, do not address the issue of whether the parallel reductions in selectivity and sensitivity are due to similar pathologie conditions or whether the two groups have distinct patterns of cochlear impairment resulting in nearly identical functional disrup- tions. Yet a third explanation consistent with their findings is that similar alterations in absolute sensitivity can pro- duce equivalent reductions in frequency resolution inde- pendent of hearing loss etiology or the presence of eochlear pathology. That is, there may be an interdependence be- tween the mechanisms mediating frequency selectivity and absolute sensitivity such that disruptions to either system produce changes in both.

The purpose of the present investigation was twofold. First, we wanted to determine whether there are age- related changes in auditory-filter shapes in frequency re- gions known to be important for speech perception. Al- though Peters and Moore (1992b) demonstrated that auditory filters can remain unchanged in elderly listeners at low center frequencies, where absolute sensitivity is not affected by presbyeusis, there is little indication that similar results obtain for higher frequency regions. Determining

whether there are age-related changes in auditory filters at higher frequencies is important for assessing whether re- duced frequency resolution is an important contributor to the poorer-than-expected speech-in-noise performance of elderly listeners. Furthermore, given the equivocal results of Peters and Moore (1992a) and Peters et al. (1992) re- garding age-related changes in processing efficiency, we wanted to evaluate whether older listeners exhibit a re-

dueed capacity to detect signals following auditory filter- ing. A second reason for conducting this study was to de- termine whether cochlear pathology is a necessary condition for producing changes in auditory filters or whether elevated thresholds alone are sufficient to cause

reductions in frequency selectivity. To accomplish these goals, auditory-filter shapes were measured in four groups of subjects: (1) young normal-hearing listeners; (2) eld- erly normal-hearing listeners; (3) elderly hearing-impaired subjects; and (4) normal-hearing young subjects with hearing losses simulated by noise masking.

This last group, the simulated-loss condition, consisted of young normal-hearing subjects tested with an additional narrow-band masker centered at the signal frequency. The purpose of the additional masker was to elevate thresholds in a range of frequencies at least a critical-band wide such that young and elderly subjects had similar hearing levels in the spectral region near the center frequency of the au- ditory filter being tested. Additionally, simulating a hear- ing loss in normal-hearing individuals provides a means of evaluating auditory function under conditions in which dy- namic ranges and loudness recruitment more closely par- allel that observed in hearing-impaired listeners.

Testing subjects with simulated hearing losses, how- ever, is useful only if the introduction of externa noise to an otherwise intact cochlea provides a valid model of the psychoacoustie capabilities of hearing-impaired subjects. Humes et al. (1988) addressed this issue by developing a model for predicting the performance of noise-masked normal-hearing listeners in a variety of psyehoaeoustic tasks. The model assumes that thresholds for subjects with simulated hearing-losses are identical to the quiet thresholds of hearing-impaired listeners. Humes etal. (1988) evaluated the model retrospectively by deter- mining whether the model's predictions for performance in simulated-loss conditions were in reasonable agree- ment with data obtained from listeners with actual hearing impairments. Their results demonstrated that, for a number of measures of frequency resolution, including auditory-filter shapes, the predicted data for noise- masked normal subjects agreed quite well with thresh- olds obtained from hearing-impaired subjects with comparable threshold elevations. These findings suggest that noise-masked normal listeners can provide appro- priate controls for determining the extent to which changes in certain psychoacoustic capacities are due to cochlear pathology or elevated thresholds. In the present investiga- tion, the two comparisons that are of particular importance are measures of auditory-filter shapes in young noise- masked normal-hearing subjects versus elderly hearing- impaired listeners and normal-hearing young subjects

2904 J. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2904

TABLE I. Absolute thresholds, in dB HL, for the test ear of elderly subjects.

Frequency (kHz) Subject 0.25 0.5 I 2 4 8

DN 25 15 20 5 35 40

RA 10 15 20 15 55 70

CR 15 15 5 15 15 35

PY 25 20 10 20 35 70

MR 20 25 30 30 50 60

FR 40 25 25 25 75 85

GH 5 20 10 25 40 90

MF 5 5 20 45 50 50

JW 25 40 40 40 40 65

DL 15 30 45 45 65 60

JK 35 35 45 50 90+ 90+

versus normal-hearing elderly individuals. Similar auditory-filter shapes for these two comparison groups would be consistent with the hypothesis that age per se does not alter the shape of auditory filters and that com- parable threshold elevation, rather than specific cochlear pathology, is sufficient to produce corresponding changes in auditory-filter shapes.

I. METHODS

A. Subjects

Eleven elderly (age 68-83) and four young (age 25- 33) listeners served as experimental subjects. The older listeners all displayed some degree of high-frequency hear- ing impairment but differed with respect to the configura- tion of their losses. Table I lists pure-tone air-conduction hearing levels (ANSI, 1989) for the test ear of elderly listeners at octave frequencies between 0.25 and 8 kHz.

Elderly subjects were divided into three groups based on the extent of their hearing loss at 2 kHz. The mean absolute threshold at 2 kHz for the four young normal- hearing subjects was taken as 0 dB HL and hearing losses for elderly listeners were established relative to this value. The first group, subsequently referred to as normal-hearing elderly listeners, consisted of three older subjects with 2-kHz thresholds less than 15 dB HL. Group two, the elderly mild impairment group, contained four older lis- teners with absolute thresholds between 20 and 30 dB HL.

The final group, referred to as elderly moderate impair- ment, had four listeners with thresholds between 40 and 50 dB HL at 2 kHz. The four younger listeners all had 2-kHz thresholds that were less than 10 dB HL (ANSI, 1989) and reported no history of otologic problems. Each of the younger subjects was tested in quiet and in the presence of two levels of a narrow-band masker designed to elevate thresholds at 2 kHz by either 20 or 40 dB. One of the younger subjects was author MS. All subjects had previous experience in psychoacoustic experiments and were paid an hourly rate.

B. Stimuli and equipment

Auditory-filter shapes were measured using the notched-noise procedure developed by Patterson (1976)

and later modified by Glasberg and Moore (1990). Two separate noise bands were positioned either symmetrically or asymmetrically around a pure-tone signal (e.g. f•). Thresholds for detecting the signal were measured as a function of the spectral separation, or notch, between the two bands. The noise bands were generated by independent sources (General Radio, 1390B and Grason Stadler, 901B ) and then passed through two cascaded low-pass filters (Kemo VBF/25) with cutoff frequencies of 400 Hz and slopes of 135 dB/octave.

After filtering, each band was multiplied by separate sinusoids which shifted the center frequency and increased the slope of the skirts. Each of the resulting noise bands was 800 Hz wide and had skirts that fell off at 270 dB/

octave. The spectrum level of the noise was set to 50 dB (re: 20/•Pa). In the symmetric condition, the noise bands were positioned above and below fs such that, expressed relative to the signal frequency, the deviation of the nearer edge of each band (A f) fromf• [i.e., (Af/•)] was 0.0 (no notch), 0.05, 0.1, 0.2, 0.3, 0.4, or 0.5. In one set of asym- metric conditions, the nearer edge of the noise band below • was closer to the signal with (Af)/• for this band set at 0.05, 0.1, 0.2, 0.3, or 0.4. The nearer edge of the masker above the signal was 0.2 farther away (i.e., Af/f• for the lower and upper bands, respectively, were 0.05, 0.25; 0.1, 0.3; 0.2, 0.4; 0.3, 0.5; and 0.4, 0.6). The remaining set of asymmetric conditions were mirror images of the first with the nearer edge of the upper band closer to des and the nearer edge of the lower band 0.2 farther away. The two narrow-band maskers were added together, electronically gated (TTES) with 20-ms cosine squared ramps, amplified (Crown D75), and presented during both intervals of a two-alternative forced choice paradigm through insert phones (Etymotic ER-3A). Sound-pressure levels refer to those generated in an HA-2, 2-cm 3 coupler.

The signal was a 2-kHz pure tone produced by a func- tion generator (Wavetek 185). It was 400 ms in duration and was gated on and off with 20-ms cosine squared ramps. Signal level was controlled by a programmable attenuator (TTES). The signal was mixed with the notched-noise masker prior to amplification and transduction.

C. General procedure

Subjects were tested individually while seated in a double-walled sound attenuating booth (IAC). Masked thresholds were determined using an adaptive two-interval forced-choice (2IFC) procedure. Each interval was 500 ms in duration and was marked on a CRT screen coincident

with stimulus presentation. There was a 500-ms silent pe- riod between the two intervals. The signal-was temporally centered within one of the intervals selected randomly on each trial. The order of masker conditions (notch widths) was also randomly determined for each subject. For each notch width, the signal was initially set at least 10 dB above the subject's estimated threshold. The signal was decreased 10 dB following two consecutive correct detec- tions and increased 10 dB after the first miss. The step size was reduced from 10 to 2 dB after the first incorrect re-

sponse.

2905 J. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2905

A change from either a correct detection to a miss or vice-versa defined a reversal. Threshold was determined as

the mean signal level for the final 8 of 11 reversals. Subjects were tested on a given condition until the standard devia- tion of three consecutive threshold measures was less than

3 dB. This same procedure, without the notched noise, was used to determine pure-tone air-conduction thresholds at 2 kHz.

As noted above, to dissociate the effects of age and hearing impairment on the shape of auditory filters, sub- jects were divided into three groups according to the extent of their hearing loss (either simulated or actual) relative to the mean 2-kHz threshold of the normal-hearing young listeners. Within each group, masked thresholds were av- eraged separately for young and elderly listeners and auditory-filter shapes were derived using these mean threshold values.

D. Auditory-filter shapes in noise-masked normal- hearing listeners

For the four young normal-hearing subjects, ntched- noise thresholds were also obtained in the presence of a narrow-band noise. The additional masker was designed to elevate thresholds in the young subjects such that absolute sensitivity (and, presumably, loudness recruitment) was similar to that of elderly listeners. The bandwidth of the noise used to simulate hearing losses was 150 Hz (a value exceeding average critical band measures at 2 kHz in normal-hearing subjects) and was centered at 2 kHz. It was produced by low-pass filtering a separate noise source at 75 Hz with two cascaded 135-dB/octave filters and mul-

tiplying the output of the filters by a 2-kHz sinusold. The resulting noise band was varied in amplitude as necessary to match the masked thresholds of young listeners to the absolute thresholds of the elderly.

The normal-hearing young listeners were tested with two levels of the narrow-band noise sufficient to elevate

quiet thresholds at 2 kHz by 20 and 40 dB. The narrow- band masker was digitally recorded (Panasonic DAT SV3500} and played continuously during simulated hear- ing loss conditions. To minimize interactions between the noise used to produce hearing losses and the noise bands from the notched-noise masker, the no-notch condition was eliminated when testing subjects with simulated hear- ing losses. Thus only when the relative deviation of either the upper or lower band of the notched-noise masker was 0.05f• was there both spectral and temporal overlap be- tween the masker used to derive auditory-filter shapes and the narrow-band noise used to simulate hearing losses.

E. Derivation of auditory-filter shapes

Auditory-filter shapes were derived based on models developed by Patterson (1976) and Glasberg and Moore (1990). According to these models, filter shapes are deter- mined by fitting smooth-line functions to the threshold data and taking the derivative of this function. The re-

sponse of the filter at a given frequency is therefore pro- portional to the slope of the threshold function at that frequency.

In general, auditory filters derived using this proce- dure take the form of two exponential functions placed side by side with a rounded top. Patterson et al. (1982) termed this the rounded exponential (roex) filter and showed that it can be well fit using two parameters: • p which defines the filter's passband and r which limits its dynamic range. The shape of the filter W(g) is given by

W(g) = (1 --r) ( 1 +pg)exp(--pg) +r, ( I )

where W is the intensity weighting function of the filter and g is the normalized distance from the filter's center frequency. The parameter p defines the slope of the filter's skirts and is allowed to differ for the upper and lower halves of the filter (the slope of the lower skirt is given by Pt and the slope of the upper skirt is given by P,). The value of r places a limit on the dynamic range of the filter and is assumed to be equal for the upper and lower halves. A parameter K, which provides an estimate of the effi- ciency of signal detection after filtering, can also be ob- tained using the roex model. Here, K indicates the signal- to-noise ratio at the output of the filter required to reach threshold (70.7% correct using the present paradigm). The value of K is reflected in the absolute levels of individ-

ual threshold curves.

Filter shapes were derived by (1) assuming starting values for P, and Pt; (2) calculating threshold curves for these values from Eq. ( 1 ) (for both symmetric and asym- metric conditions); (3) determining the mean-squared de- viation between predicted and obtained thresholds; and (4) altering filter parameters so as to minimize the mean- squared deviation between actual and predicted threshold curves. These procedures were implemented using a com- puter program published by Glasberg and Moore (1990). It was assumed that, for a given notch width, the center frequency of the filter used to detect the 2-kHz signal, was the one yielding the highest signal-to-noise ratio. The pro- gram used to derive auditory-filter parameters allowed for shifts in the center frequency equivalent to 0.15rs and al- tered the bandwidth of the filters as described by Glasberg and Moore (1990) to account for changes in filter shape as a function of center frequency.

II. RESULTS

A. Auditory-filter shapes and age

Figures 1 and 2 show average masked thresholds and auditory-filter shapes for the normal-hearing (as defined above) young and elderly subjects. Figure I displays aver- age thresholds as a function of relative notch width [(A f/ f•)]. The asterisks are for conditions in which the notch was positioned symmetrically about the signal. Squares show thresholds obtained when the noise band above the

signal was 0.2 farther away and triangles show thresholds obtained when the lower frequency band was 0.2 farther away. Differences between these latter two conditions in- dicate the degree of asymmetry within the auditory filter.

2906 d. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2906

Young normal-hem'ing Elderly normal-hearing

60 *

Rflafivc •:viafi0n of n•rrx cdgc Rdalivc dcviali0n of nc•r edge

FIG. 1. Average masked thresholds for young (left) and elderly (right) subjects with hearing levels less than 15 dB HL (relative to mean of young subjects). Asterisks show thresholds for symmetric notch configurations. Squares give thresholds for asymmetric notch conditions in which the low-frequency band was closer to the signal. Triangles show thresholds for asymmetric conditions with the high-frequency band closer to the test frequency. The solid lines show data predicted by the rocx (p,r) filter model. Points labeled as "Q" are average air-conduction quiet thresholds.

Points labeled as "Q" on each graph represent the average quiet threshold in dB SPL for the two groups. In both the symmetric and asymmetric conditions, thresholds in quiet were below those measured for the widest notch width

indicating that the masking noise, rather than absolute sen- sitivity, was limiting detection. Solid lines show thresholds predicted by the roex (p,r) model. The average root-mean- squared (rms) deviation of obtained and predicted thresh- olds was 1.18 dB for young listeners and 1.61 dB for older subjects. Overall, threshold curves for the young and eld- erly listeners were quite similar. Both groups exhibited the highest thresholds for conditions in which the notch was placed symmetrically about the signa and the lowest thresholds for asymmetric conditions in which the upper band was 0.2 closer than the lower one. The principal dif-

o I ¾ou.g

• -xo

o

• -20

-.5 -.s. -.4 -.3 -.2 -.I 0 .I 2 3 .4

No•ali• deviation from center

FI•. 2. Auditor-filter sha• for young (solid line) •d elderly (d•h• line) no•-h•ng listcne• dedv• fro• the thruhold data •own in Fig. 1.

ference between the masking data for young and elderly normal-hearing subjects was that there was approximately a 5-dB greater drop in thresholds for the young listeners between the smallest and largest notch widths.

The similarity in masked thresholds is reflected in the auditory-filter shapes derived from these data which are shown in Fig. 2. The dashed line indicates the filter for the older group while the solid curve shows the corresponding filter for younger subjects. Except for a 5- to 7-dB smaller dynamic range for elderly listeners, there was considerable overlap in the shape of auditory filters for the two groups. The efficiency of signal detection after filtering (K) is rep- resented by the relative vertical positions of the two func- tions. The filter for elderly listeners was not displaced ver- tically relative to that for younger subjects suggesting that processing efficiency did not differ between the two groups.

Table II lists individual filter-shape parameters for all subjects tested in this study. Column headings, except the final one, correspond to parameters defined above. The last column, labeled "Asymm," is a measure of the degree of asymmetry within the filter and is represented by the value PulP1. Comparison of values for the first two groups listed in the table, the young and elderly normal-hearing listen- ers, indicated that there were no significant differences in filter-shape parameters as a function of age (Student's t test for differences between young and elderly for Pa Pu, ERB, K, r, and Asymm were: --0.65, --0.55, --0.53, --0.02, --0.74, --0.22, respectively; in all cases p>0.05). The largest ERB value obtained for any of the normal-hearing listeners was 0.165 for subject MS who was part of the young listener group.

Figure 3 shows average masked thresholds for subjects with mild hearing losses (either young subjects with sim- ulated 20-dB hearing losses or elderly subjects with 20- to 30-dB hearing impairments). As was the case for normal-

2907 J. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2907

TABLE II. Auditory-filter shape parameters for young normal-heating, elderly normal-hearing, young noise-masked and elderly hearing-impaired listeners. Column headings are: HL hearing level (relative to the average of the four normal-hearing young subjects); Pt•slope of low-frequency side; Pu--slope of high-frequency side; ERB•equivalent rectangular bandwidth; r•dynamie range; K---efficiency measure; Asymm--asymmetry of the filter expressed as P•/Pt.

Subject HL(dB) Pt Pu ERB(kHz) r K Asymm

Normal-hearing young LC 0 24.1 34.1 0.141 64.8 3.9 1.41 LL 0 23.7 32.0 0.147 56.4 7.2 1.35 CC 0 21.0 30.3 0.161 87.9 4.7 1.44 MS 0 20.9 28.7 0.165 88.8 9.9 1.37

Normal-hearing elderly

Simulated mild loss

Simulated moderate loss

Elderly mild impairment

Elderly moderate impairment

CR 14 20.0 33.3 0.160 88.3 9.0 1.66 DN 5 28.2 22.2 0.161 56.3 3.3 0.78 RA 15 20.9 34.8 0.153 47.5 7.1 1.66

LC 20 25.4 32.2 0.141 56.9 5.3 1.26 LL 20 22.0 31.5 0.154 51.5 10.0 1.43 CC 20 19.7 26.3 0.178 51.7 3.5 1.33 MS 20 16.9 21.7 0.210 44.0 0.7 1.28

LC 40 15.6 21.5 0.222 37.1 0.7 1.37 LL 40 15.1 16.7 0.252 35.1 5.7 1.10 CC 40 13.8 15.2 0.277 36.0 1.3 1.10 MS 40 8.0 11.7 0.422 41.5 - 2.5 1.46

PY 18 16.5 29.8 0.188 70.9 5.8 1.80 FR 21 17.5 25.6 0.192 80.7 5.8 1.46 GH 21 16.5 31.0 0.186 68.0 7.8 1.87 MR 29 13.1 50.0 0.193 42.3 2.4 3.80

MF 41 20.3 15.3 0.229 37,6 -0.6 0.75 JW 41 10.0 25.2 0.279 62.1 6.4 2.52 DL 42 8.8 45.1 0.271 24.7 1.1 5.12

JK 49 6.8 38.9 0.344 26.5 5.7 5.72

hearing listeners, threshold curves for the two groups were quite similar. The principal distinction between masked thresholds for older and younger subjects was that the eld- erly listeners demonstrated greater differences between the two asymmetric conditions. The data were well fit by the roex (p,r) model with average deviations between pre- dicted and obtained thresholds of 1.22 and 1.48 dB for

younger and older subjects, respectively. Thresholds for the widest notch widths in the asymmetric condition with the high-frequency band closer to fs (triangles) ap- proached subjects' absolute sensitivity at the test frequency but in all cases masked thresholds remained above hearing levels in quiet.

Filter shapes for young and elderly subjects with mild hearing losses are shown in Fig. 4. The greater asymmetry for older listeners, noted in the threshold data, was due largely to a shallower low-frequency slope for these sub- jects. However, values of Pt were not significantly different (t=1.98; p>0.05) for young and elderly listeners. The parameter Asymm, which measured overall filter asymme- try, did show significant age-related changes (t=2.71; p<0.05) for the two groups suggesting slightly more asymmetrical filters for the elderly listeners. Differences between younger and older subjects on the remaining pa- rameters Pu, ERB, K, and r were all nonsignificant (t ----0.12, --1.09, --0.17, and 1.27, respectively; for all com- parisons, œ > 0.05).

Figures 5 and 6 show masked thresholds and auditory- filter shapes for listeners with moderate hearing losses

(40-dB simulated loss in young or 40- to 50-dB actual loss in elderly). The roex (p,r)model again provided a good fit to the data with average rms deviations between predicted and obtained thresholds for younger and older listeners of 1.01 and 1.25 dB, respectively. Both groups of subjects exhibited flatter threshold curves and broader auditory- filter shapes compared to listeners with either normal hear- ing or mild losses. However, the ERBs of filters for young and elderly subjects in the moderate-loss condition were not significantly different (t=0.25; p>0.05). Further- more, as was the case for listeners with mild hearing im- pairments, auditory filters were more asymmetrical (t =- 2.96; p < 0.05) for elderly subjects with moderate im- pairments than their younger noise-masked controls. This difference in filter asymmetry was due to combined effects of both the high- and low-frequency sides because there was no significant difference between the two age groups with respect to either slope alone (t for differences between young and elderly for P/=0.25; for Pu=0.96; in both cases p> 0.05). Other filter parameters including the dynamic range and detection efficiency did not differ as a function of age (t for r=0.03 and for K= --0.77; p>0.05).

Figure 7 summarizes the relationship between ERB and hearing level for all conditions examined in the present study. Open symbols indicate filter widths for subjects with hearing impairments while closed symbols show ERBs for normal-hearing listeners. Squares show data for young sub- jects and triangles display findings for older individuals. The figure illustrates that, for both young and elderly sub-

2908 J. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2908

Young (simulated 20-dB loss) Elderly (20-30 dB HL)

0,1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

Relative devhfion of nearer cdgc Relative deviation of near edge

FIG. 3. Same as Fig. I except data are for young subjects with 20-dB simulated hearing losses (left) and elderly listeners with thresholds between 20 and 30 dB HL (right).

jects, ERBs were similar for hearing levels less than 15-20 dB after which they increased with increasing threshold elevation. Thus, for a given hearing level, there was con- siderable overlap between the ERBs of younger and older subjects. The data were best fit by an exponential function of the form:

ERB=0.016+0.14(absolute threshold). (2)

B. Relationship among auditory filter parameters

Table III displays Pearson product-moment correla- tions between auditory filter parameters for subjects tested in quiet (i.e., data from simulated hearing losses have been excluded). As expected, both age and hearing loss were significantly correlated with the ERBs of auditory filters. However, age was also significantly correlated with hearing

0 [ Young I

• -•o

'• -20

-30

-.6 -.5 -.4 -.3 -.2 -.l 0 .1 .2 .3 .4 .5 6

Normalized deviation from center frequency

FIG. 4. Auditory filter shapes derived from masking data of Fig. 3. The solid line shows filter shapes for young listeners; the dashed line shows filters for elderly subjects.

level suggesting that the two variables, age and hearing loss, may not independently affect auditory-filter widths. Furthermore, there was a stronger correlation between hearing level and ERBs (0.91) than between age and ERBs (0.51 ) suggesting that changes in absolute sensitiv- ity may be the principal contributor to increases in the width of auditory filters.

Table IV shows correlations between auditory-filter shape parameters with the effects of hearing level partialed out. As indicated in the table, once differences in absolute sensitivity were removed from the correlations, there was no longer a significant relationship between age and ERB. The only correlations that remained significant after the effects of absolute threshold had been partialed out were between ERB and asymmetry and between dynamic range and efficiency.

C. Comparison of simulated and actual hearing losses

To determine whether simulated and actual hearing impairments have comparable effects on auditory-filter shapes, filter parameters for hearing-impaired and masked normal-hearing listeners were compared for both the mild and moderate hearing loss groups. Only one parameter, Asymm, which measures the degree of asymmetry within the filter, was significantly different as a function of actual, as opposed to simulated, hearing loss (t for: mild hearing loss group = --2.97; moderate hearing loss group = --2.17; p<0.05 for both groups). ERB (t=-0.26; p>0.05), K (t = -- 0.97; p > 0.05) and r (t = -- 0.28; p > 0.05) were not differentially affected by simulated or actual hearing losses.

III. DISCUSSION

This study compared auditory-filter shapes in young normal-hearing, elderly normal-hearing, elderly hearing- impaired, and young noise-masked listeners. Young and

2909 d. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2909

Young (simulated 40-dB loss) Elderly (40-50 dB HL)

I0 I I I I I I ' I I I I I I o o.l 0.2 o3 0.4 o3 o o.I 0.2 0.3 0.4 03

Relative deviation of nearer edge Relative deviation of nearer edge

FIG. 5. Same as in Fig. I except thresholds are for young listeners with simulated 40-dB heating losses (left) and elderly listeners with thresholds between 40 and 50 dB HL (right).

elderly listeners with hearing thresholds less than 20 dB HL at 2 kHz exhibited similar auditory-filter shapes. Fur- thermore, comparable threshold elevations, independent of age, produced similar changes in filter shape parameters. Comparison of these findings with previous measures of frequency selectivity in both normal-hearing and hearing- impaired populations is somewhat difficult, however, ow- ing to a number of methodological differences between ex- periments.

One study (Glasberg and Moore, 1986) measured auditory-filter shapes at 2 kHz in unilaterally and bilater- ally heating-impaired listeners under conditions nearly identical to those of the present investigation. Excluding data from the subject with the highest masked thresholds, they found average values for ERB, Pt, and Pu in unim- paired ears of 0.177, 19.0, and 29.25, respectively. Mean values for these parameters in normal-hearing subjects

(both young and elderly) in the present experiment were O. 155, 22.6, and 30.7. Furthermore, the hearing loss in five ears of their bilaterally hearing-impaired subjects was be- tween 40 and 45 dB. The average value for ERB, P/, and Pu in these five ears was 0.286, 11.8, and 23.0, respectively. Corresponding results from subjects with 40- to 50-dB sim- ulated or actual hearing losses in the current experiment showed ERB, P/and P• values of 0.287, 12.3, and 20.6. Thus data from both normal-hearing and hearing-impaired listeners in the two studies were quite similar. Other inves- tigations (Moore, 1987; Glasberg and Moore, 1990), how- ever, have found ERB values for normal-hearing subjects that were somewhat higher than those reported in the present study. Glasberg and Moore (1990), for instance, summarized the results of a number of auditory-filter shape experiments (Moore and Glasberg, 1983; Dubno and

•o 40

.,•

-30

-40 I { I I I I I I I I I I I -.6 -.5 -4 -.3 -.2 -.1 0 . I .2 .3 .4 5 6

Normalized deviation from center frequency

FIG. 6. Auditory filter shapes derived from masking data of Fig. 5. Solid and dashed lines are for young and elderly subjects, respectively.

500

4O0

2OO

Normal-H•ring young Normal-He•ring elderly

Simulated hoearing loss Actual h•ring loss

[]

I ß ß [] []

OAA 0

loo I , I , I . , I I , 0 10 20 30 40 50

Hearing Level (clB)

FIG. 7. ERB as a function of hearing level for normal-hearing, simulated hearing-loss, and hearing-impaired listeners. Symbol definitions are given in the legend.

2910 J. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2910

TABLE IIl. Pearson product-moment correlations between auditory- filter shape parameters. Variable names are as given in Table II. *--significant at the 0.05 level; **--significant at the 0.01 level.

Age ERB K r Asymm HL

Age 0.51' -0.22 0.37 0.28 0.69** ERB -0.29 0.63* 0.82** 0.91'*

K -0.61' -0.12 -0.41

r 0.61' 0.66*

Asymm 0.67** HL

Dirks, 1989; Shailer et al., 1990; Moore et al., 1990) which employed notched-noise maskers and moderate to high spectrum levels. The equation they found that provided the best fit for ERB values over the frequency range 0.1 to 10 kHz was

ERB = 24.7 (4.37F+ 1 ), (3)

where F is frequency in kHz. For a filter centered at 2 kHz, this equation gives a predicted ERB value of approxi- mately 0.24 kHz which is somewhat higher than the aver- age ERBs obtained from normal-hearing subjects in the present experiment. The discrepancy may be due, at least in part, to the fact that Eq. (3) was obtained from studies employing a number of different procedures for deriving auditory-filter shapes. For example, some of the studies employed conditions in which the notch was placed asym- metrically about the signal while others only included sym- metric conditions. Furthermore, only one of the studies (Moore et al., 1990) adjusted filter bandwidths to correct for shifts in the center frequency that can occur when lis- teners attempt to optimize signal-to-noise ratios. As Glas- berg and Moore (1990) have demonstrated, both of these factors, inclusion of asymmetric conditions and compen- sating for shifts in filter center frequency, can have sub- stantial effects on the shape of auditory filters. Despite these differences, however, there is overall agreement be- tween the studies with respect to the general shape of au- ditory filters in normal-hearing subjects.

The present results also agree with recent findings (Glasberg and Moore, 1986; Peters and Moore, 1992a,b; Peters et al., 1992) regarding the effects of age on fre- quency selectivity. Specifically, they argue that previous demonstrations of age-related changes in auditory-filter shapes (Patterson et al., 1982) are likely due to confound- ing the effects of age and hearing loss on the shape of

TABLE IV. Partial correlations between auditory-filter shape parameters (the effects of hearing level have been removed). Variable names are as given in Table II. *--significant at the 0.05 level; **--significant at the 0.01 level.

Age ERB K r Asymm

Age -0.39 0.08 --0.16 --0.32 ERB 0.20 0.09 0.68**

K --0.61'* --0.12

r 0.30

Asymm

auditory filters. [However, see Glasberg et al. (1984) and Lutman et al. (1991) for suggestions of a correlation be- tween age and frequency selectivity.] For example, al- though the study by Glasberg and Moore (1986) measur- ing auditory-filter shapes in hearing-impaired subjects was not specifically designed to assess the effects of age on fre- quency selectivity, analysis of results from subjects with unilateral losses suggests that there was no systematic change in ERBs as a function of age; listeners with unilat- eral losses ranged in age from 47-70, yet there was no significant correlation between age and ERBs of auditory filters. Rather, ERBs remained constant for hearing losses less than approximately 30-35 dB and increased linearly for hearing levels greater than this. These findings directly parallel those of the present investigation in which subjects with thresholds less than 20 dB HL ranged in age from 25-74 yet showed no differences with respect to the width of auditory filters. Peters and Moore (1992a) and Peters et al. (1992) also demonstrated that there was no signifi- cant correlation between age and ERBs for signal frequen- cies of 400 and 800 Hz [there was some suggestion in their data that at very low frequencies of 100 and 200 Hz there was a broadening of auditory-filter shapes as a function of age]. Finally, Glasberg and Moore (1989) reported a cor- relation of --0.16 between ERB and age after partialing out the effects of hearing loss. Taken together, these find- ings support the hypothesis that hearing level, rather than age, is the predominant factor affecting the shape of audi- tory filters among the elderly.

The results of this experiment are also consistent with previous research (Patterson et al., 1982; Glasberg et al., 1984; Lutman et al., 1991 ) suggesting that, for frequencies greater than 1 kHz, the efficiency of signal detection fol- lowing auditory filtering, denoted by the parameter K, does not change as a function of age. Peters and Moore (1992a), however, have reported significant negative cor- relations between age and processing efficiency for signals at octave frequencies between 100 and 800 Hz. They sug- gested that one reason this relationship may not have been found in prior investigations is that inherent fluctuations in the notched-noise masker used to measure auditory-filter shapes can impair signal detection. Since the fluctuations are more rapid, and hence easier to detect, when they are processed through broader auditory filters, previous inves- tigations may have masked any increase in K as a function of age by measuring filter shapes in elderly listeners at frequencies where ERBs were greater than normal. That is, the broader-than-normal filters of elderly subjects caused more rapid fluctuations at the output of the filter which may have made the signal easier to detect. In the study by Peters and Moore (1992a), auditory-filter shapes were measured at signal frequencies where ERB values were similar for older and younger listeners which allowed the decreased efficiency for older subjects to become apparent. The present results argue against this explanation for the similarity of processing efficiency in older and younger lis- teners at 2 kHz because ERB values for the two groups at this frequency were nearly identical. The possibility re- mains, however, that processing efficiency is decreased in

2911 J. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2911

older subjects at lower frequencies. Peters and Moore (1992a) provided initial support for this hypothesis by demonstrating that, as signal frequency is reduced from 800 to 100 Hz, there is a greater increase in K for elderly normal-hearing subjects than for unimpaired young listen- ers. That is, both groups exhibit poorer processing effi- ciency at lower frequencies where inherent noise fluctua- tions have a greater effect on performance (Moore et al., 1990). The increase in K is greater for elderly subjects, however, because they may also have reduced processing efficiency at these lower frequencies.

It should be noted that values of K for the normal-

hearing young and elderly subjects in the present study were generally higher and more variable than have been reported by other investigators using similar experimental parameters (Glasberg and Moore, 1986; Glasberg et al., 1984). Although it is not clear what might account for the higher K values in this investigation, processing efficiency was elevated equally for both elderly and young listeners (Glasberg and Moore, 1986; Glasberg el al., 1984). There- fore, the findings from this study are consistent with the suggestion that post-filter detection efficiency at 2 kHz is not reduced as a function of age. It is also worth noting that K can be quite variable even among similarly aged listeners with comparable hearing levels [for example, in the simulated mild loss condition of the present study, K values ranged from 0.7 to 10.0] and therefore comparisons across different populations must be made cautiously.

The finding that both simulated and actual hearing losses can produce equivalent reductions in frequency se- lectivity provides further empirical support for the model proposed by Humes et al. (1988) specifying how simulated hearing losses can be used to predict the effects of hearing- impairment on certain psychoacoustic tasks (principally those involving fixed-level maskers). Recall that their model (Humes et al., 1988) assumes that thresholds for simulated-loss subjects are identical to the quiet thresholds of hearing-impaired listeners. One evaluation of the model involved comparing predicted thresholds in an auditory- filter shape paradigm (for a hypothetical group of simulated-hearing loss subjects) with data obtained from hearing-impaired listeners (Tyler et al., 1984; Glasberg and Moore, 1986). Humes et al. (1988) found excellent

agreement between masking data obtained in the auditory- filter shape experiments and thresholds predicted by the model. The present results provide further support for the model by demonstrating that subjects with approximately equal simulated and actual hearing losses exhibit nearly identical changes in auditory-filter shapes. Thus masked thresholds predicted by the model for a hypothetical group of subjects with simulated hearing losses were in good agreement with those obtained using narrow-band maskers to simulate 20- and 40-dB hearing losses. More impor- tantly, both the predicted and obtained simulated-loss thresholds were not significantly different from those mea- sured in subjects with actual hearing impairments.

Dubno and Schaefer (1992) have also recently com- pared auditory-filter shapes in listeners with comparable simulated and actual hearing losses at 800, 1200, and 2000

Hz. However, unlike the present investigation, they used a spectrally shaped broadband masker to match thresholds of hearing-impaired and noise-masked listeners. Overall, the pattern of results they obtained were similar to those of the present study. At 2 kHz, for example, masked thresh- olds for hearing-impaired listeners in their investigation were 2.0 to 4.7 dB higher than those for noise-masked normal-hearing subjects. This finding compares favorably with the 2.29- to 7.91-dB differences in masked thresholds

obtained in the present study between subjects with actual and simulated mild hearing impairments (20-30 dB HL at 2 kHz). It is also in good agreement with findings from the moderate hearing-loss condition in our investigation (40-50 dB HL) in which hearing-impaired listeners exhib- ited masked thresholds that were between 0.18 and 3.16 dB

higher than the corresponding simulated-loss condition. 2 In summary, results from both this study and Dubno and Schaefer (1992) indicate that the slope of the function relating notch width to masked threshold is similar, but slightly shallower, for subjects with thresholds elevated due to cochlear pathology than for losses produced by the introduction of either wideband or narrow-band noise.

An important distinction between the results of this study and those of Dubno and Schaefer (1992), however, is that, in the present investigation, auditory-filter shape parameters were not significantly different for noise- masked normal-hearing and elderly hearing-impaired lis- teners. In contrast, Dubno and Schaefer (1992) observed significant differences for the filter-shape parameters p and ERB in listeners with actual and simulated hearing losses. The difference in the two studies may be, at least in part, attributable to differences in the maskers used to simulate

the hearing losses. In this study, normal-hearing thresholds were elevated using a narrow-band masker centered at the signal frequency. This avoided spectral overlap between the notched-noise masker used to measure auditory-filter shapes and the masker used to simulate hearing losses. Dubno and Schaefer (1992) employed a spectrally shaped broadband simultaneous masker to match thresholds in

normal-hearing and hearing-impaired subjects. Humes and •Iesteadt (1989) have demonstrated that normal-hearing listeners exhibit less combined masking when there is both spectral and temporal overlap between maskers. Thus, in the study by Dubno and Schaefer (1992), the presence of a more intense second masker in the simulated-loss condi-

tions may have attenuated the effectiveness of the broad- band noise used to simulate hearing losses, resulting in lower masked thresholds for the noise-masked normal-

hearing listeners. The effect of this reduction in masked thresholds would be to make auditory-filter parameters more like those for unimpaired listeners in quiet. Consis- tent with this hypothesis, Dubno and Schaefer (1992) found that values for ERB and p were quite similar in normal-hearing subjects tested in quiet or with broadband spectrally shaped noise.

The similarity of auditory-filter shapes in individuals with simulated and actual hearing impairments that was demonstrated in the present experiment cautions against using behavioral measures of frequency selectivity to estab-

2912 J. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2912

lish cochlear mechanisms responsible for changes in fre- quency resolution. Peters and Moore (1992a) and Peters et al. (1992), for instance, have suggested that demonstra- tions of analogous changes in auditory-filter shapes in young and elderly subjects matched for hearing loss indi- cates that both idiopathic loss in the young and presbyeusic loss in the elderly impair the active cochlear mechanisms thought to be largely responsible for the sharp tuning of the cochlea (Yates, 1986; Pickles, 1988). Furthermore, they interpret their finding (Peters and Moore, 1992b) that similar degrees of high- and low-frequency heating impair- ment produce comparable changes in frequency resolution as an indication that the active processes operate in both the apical and basal portions of the cochlea. However, such conclusions may be premature in light of the present find- ings that the ERBs of auditory filters can increase in the absence of cochlear or retrocochlear pathology. Zwicker and Schorn (1978) have also shown that cochlear pathol- ogy is not a necessary condition for producing changes in frequency selectivity.

One alteration in auditory-filter shapes that may be a direct consequence of coehlear pathology, however, is a change in the asymmetry of the filters. As noted above, for a given amount of threshold elevation, the only difference between simulated and actual hearing-impaired listeners, was that the latter demonstrated significantly greater asymmetries in auditory filters. For the two groups with elevated thresholds (mild and moderate hearing impaired subjects), there were significant positive correlations (r =0.75 and 0.58, respectively for mild and moderate amounts of threshold elevation) between age and filter asymmetry. The principal difference between older and younger subjects in the hearing-impaired groups was that the older listeners had actual hearing losses, presumably due to some type of cochlear pathology, while the younger subjects did not. Furthermore, the correlation between age and asymmetry for subjects with elevated thresholds is un- likely due to age-related changes in asymmetry because there was no correlation between age and asymmetry (r = 0.01; p > 0.05) among normal-hearing young and elderly subjects. This finding agrees with data from Glasberg and Moore (1986) who also showed no relationship between age and filter asymmetry. Taken together, therefore, these findings suggest that one effect of cochlear pathology may be to increase the asymmetry of auditory filters relative to that found for unimpaired ears.

IV. CONCLUSIONS

(1) Auditory-filter shapes were not significantly dif- ferent in young and elderly subjects with normal hearing at the test frequency of 2 kHz.

(2) Simulated hearing losses in young subjects and actual hearing impairments in elderly listeners produced comparable changes in ERBs of auditory filters. This find- ing suggests that hearing loss, independent of specific co- chlear pathology, is sufficient to alter frequency selectivity.

(3) The efficiency of signal detection at 2 kHz was not significantly different for young and elderly subjects with similar hearing levels.

(4) Auditory filters that are altered by threshold ele- vations attributable to cochlear pathology may be more asymmetrical than those altered by the introduction of ex- ternal noise.

ACKNOWLEDGMENTS

The authors wish to thank Judy Dubno and Robert Peters for their helpful comments in reviewing earlier ver- sions of this manuscript. In addition, David Pisoni offered several important suggestions regarding the experimental design used in this study. This research was supported by Grants DC-00012-12 and DC-00012-13 from NIH

NIDCD and Grant AG08293 from NIA.

•Patterson et al. (1982) also suggested a three-parameter model which is theoretically more accurate for fitting auditory filter shapes than the two-parameter model used in the present study. This model, known as the roex (p, to, t) model, suggests that each side of the auditory filter can be approximated by the sum of two rounded exponential functions. The parameter p defines the passband and slope of the first function while •o and t define the second. Here, Wrepresents the point at which the second function takes over from the first while t represents the rate of fall of this second function. Although this provides a more accurate characteriza- tion of the auditory filter, the range of notch widths used in this exper- iment was not sufficient to allow accurate determination of u• and t.

2Dubno and Schaefer (1992) reported data only for conditions in which the notch was placed symmetrically about the signal. The above com- parisons are therefore limited to thresholds from the present experiment that were obtained with symmetrical notch widths.

ANSI (1989). ANSI S3.6-1989, "American National Standard Specifica- tion for Audiometers" (American National Standards Institute, New York).

Corso, J. F. (1971). "Sensory processes and age effects in normal adults," J. Gerontol. 26, 90-105.

Dubno, J. R., and Dirks, D. D. (1989). "Auditory filter characteristics and consonant recognition for hearing-impaired listeners," J. Acoust. Soc. Am. 85, 1666-1675.

Dubno, J. R., and Schaefer, A. B. (1992). "Comparison of frequency seleclivity and consonant recognition among hearing-impaired and masked normal-hearing listeners," J. Acoust. Soc. Am. 91, 2110-2121.

Dubno, J. R., Dirks, D. D., and Morgan, D. E. (1984). "Effects of age and mild hearing loss on speech recognition in noise," J. Acoust. Soc. Am. 76, 87-89.

Evans, E. F. (1975). "Normal and abnormal functioning of the coehlear nerve," Syrup. Zool. Soc. London 37, 133-165.

Fletcher, H. (1940). "Auditory patterns," Rev. MOd. Phys. 12, 47-65. Florentine, M., Buus, S., Scharf, B., and Zwicker, E. (1980). "Frequency

selectivity in normally-hearing and hearing-impaired observers," J. Speech Hear. Res. 23, 646-669.

Glasberg, B. R., and Moore, B.C. J. (1986). "Auditory filter shapes in subjects with unilateral and bilateral cochlear impairments," J. Acoust. Soc. Am. 79, 1020-1033.

Glasberg, B. R., and Moore, B.C. 1. (1989). "Psychoaeoustic abilities of subjects with unilateral and bilateral cochlear hearing impairments and their relationship to the ability to understand speech," Scand. Audioi. Suppl. 32, 1-25.

Glasberg, B. R., and Moore, B.C. J. (1990). "Derivation of auditory filter shapes from notched-noise data," Hear. Res. 47, 103-138.

G!asberg, B. R., Moore, B. C..I., Patterson, R. D., and Nimmo-Smith, I. (1984). "Dynamic range and asymmetry of the auditory filter," J. Acoust. Soc. Am. 76, 419-427.

Hannley, M., and Dorman, M. F. (19fi3). "Susceptibility to intraspeech spread of masking in listeners with sensorineural hearing loss," J. Acoust. Soc. Am. 74, 40-51.

Hinchcliffe, R. (1962). "The anatomical locus of presbycusis," J. Speech Hear. Disoral. 27, 301-310.

Humes, L. E., and Jesteadt, W. (1989}. "Models of the additivity of masking," J. Acoust. Soc. Am. 85, 1285-1294.

2913 J. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2913

Humes, L. E., Espinoza-Varas, B., and Watson, C. S. (1988). "Modeling sensorineural hearing loss. I. Model and retrospective evaluation," J. Acoust. Soc. Am. 83, 188-202.

Kalikow, D. N., Stevens, K. N., and Elliot, L. L. (1977). "Devdopment of a test of speech intelligibility in noise using sentence materials with controlled word predictability," J. Acoust. Soc. Am. 61, 1337-1351.

Lutman, M. E., and Clark, J. (1986). "Speech identification under sim- ulated hearing-aid frequency response characteristics in relation to sen- sitivity, frequency resolution and temporal resolution," J. Aeonst. Soc. Am. 80, 1030-1040.

Lutman, M. E., Gatehouse, S., and Worthington, A. G. (1991). "Fre- quency resolution as a function of hearing threshold level and age," J. Acoust. Soc. Am. 89, 320-328.

Moore, B.C. 3. (1987). "Distribution of auditory-filter bandwidths at 2 kHz in young normal listeners," J. Acoust. Soc. Am. 81, 1633-1635.

Moore, B.C. J., and G!asberg, B. R. (1983). "Suggested formulae for calculating auditory-filter bandwidths and excitation patterns," J. Acoust. Soc. Am. 74, 750-753.

Moore, B.C. J., Peters, R. W., and Glasberg, B. R. (1990). "Auditory filter shapes at low center frequency," J. Acoust. Soc. Am. 88. 132-140.

Patterson, R. D. (1976). "Auditory filter shapes derived with noise stim- uli," J. Acoust. Soc. Am. 59, 640-654.

Patterson, R. D., Nimmo-Smith, I., Weber, D. L., and Milroy, R. (1982). "The deterioration of hearing with age: Frequency selectivity, the crit- ical ratio, the audiogram, and speech thresholds," J. Aeoust. Soc. Am. 72, 1788-1803.

Peters, R. W., Moore, B.C. J., and Pillsbury, H. C. (1992). "Aging, auditory functions and speech discrimination," Paper presented at NIH Symposium, Aging: The Quality of Life, Washington, DC.

Peters, R. W., and Moore, B.C. $. (1992a). "Auditory filters and aging: Filters when audiometric thresholds are normal," in Auditory Physiol- ogy and Perception (Pergamon, Paris).

Peters, R. W., and Moore, B.C. J. (1992b). "Auditory filter shapes at low center frequencies in young and elderly hearing-impaired subjects," J. Acoust. Soc. Am. 91, 256-266.

Pickles, J. O. (1988)..4n Introduction to the Physiology of Hearing (Ac- ademic, London), 2nd ed.

Plomp, R., and Mirapen, A.M. (1979). "Speech-reception threshold for sentences as a function of age and noise level," $. Acoust. Soc. Am. 66, 1333-1342.

Robertson, D., and Manley, (3. A. (1974). "Manipulation of frequency analysis in the cochlear ganglion of the guinea pig," J. Comp. Phys. 91, 363-375.

Shailer, M. J., Moore, B.C. J., Giasberg, B. R., Watson, N., and Harris, S. (1990). "Auditory filter shapes at 8 and 10 kHz," J. Aeonst. Soc. Am. 88, 141-148.

Tyler, R. S. (1986). "Frequency resolution in hearing-impaired listen- ers," in Frequency Selectioiry in Hearing, edited by B.C. J. Moore (Academic, London).

Tyler, R. S., Fernandes, M., and Wood, E. J. (1980). "Masking, temporal integration, and speech intelligibility in listeners with noise-induced hearing loss," in Disorders of Auditory Function III, edited by I. Taylor and A. Markides (Academic, London).

Tyler, R. S., Hall, •. W., Glasberg, B. R., Moore, B.C. J., and Patterson, R. D. (1984). "Auditory filter asymmetry in the hearing impaired," J. Aeonst. Soc. Am. 76, 1363-1368.

Tyler, R. S., Wood, E. J., and Fernandes, M. (1982). "Frequency reso- lution and hearing loss," Brit. $. Audiol. 16, 4563.

Yates, O.K. (1986). "Frequency selectivity in the auditory periphery," in Frequency Selectioity in Hearing, edited by B.C. J. Moore (Aca- demic, London).

Zwicker, E., and Sehorn, K. (1978). "Psychoacoustical tuning curves in audiology," Audiology 17, 120-140.

2914 J. Acoust. Soc. Am., Vol. 93, No. 5, May 1993 M.S. Sommers and L. E. Humes: Auditory filter shapes 2914