Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2013 May 1.
Published in final edited form as: Ear Hear. 2012 May;33(3):411–420. doi: 10.1097/AUD.0b013e31823d78dc

Gender identification in younger and older adults: use of spectral and temporal cues in noise-vocoded speech1

Kara C Schvartz 1, Monita Chatterjee 1
PMCID: PMC3340495  NIHMSID: NIHMS339617  PMID: 22237163

Abstract

Objective

The aim of the current study was to investigate potential effects of age on the ability of normal-hearing (NH) adult listeners to utilize spectral and temporal cues when performing a voice gender identification task.

Design

Ten younger and ten older NH adult listeners were measured on their ability to correctly identify the speaker gender of six different vowel tokens (H-/vowel/-D) when spoken by eight speakers (four male, four female). Spectral (number of channels) and temporal cues (low-pass cut-off frequency for temporal envelope extraction) were systematically manipulated using noiseband vocoding techniques; stimuli contained 1, 4, 8, 16 or 32 spectral channels, while the low-pass cutoff frequency of the temporal envelope filter was 20, 50, 100, 200 or 400 Hz. Further, the fundamental frequencies (F0s) of the vowel tokens were manipulated to create two conditions: ‘Expanded’ (large range of F0 values), and ‘Compressed’ (small range of F0 values).

Results

In general, younger listeners performed better than the older listeners, but only when stimuli were spectrally degraded. For both the Expanded and Compressed conditions, the overall performance of the younger listeners was better than that of the older listeners, suggesting age-related deficits in both spectral and temporal processing. Furthermore, a significant interaction between age group and temporal envelope cues revealed that older listeners received less benefit from increasing temporal envelope information compared to the benefit observed among younger listeners. Specifically, the performance of the younger NH group (collapsed across number of channels), but not the older NH group, improved as the temporal envelope cut-off frequency was increased from 50 to 400 Hz.

Conclusions

The results reported here support previous findings of senescent declines in perceiving spectrally reduced speech and temporal amplitude modulation processing. These results suggest that when F0 values are similar to one another, younger listeners can use temporal cues alone to glean voice-pitch information but older listeners exhibit a lessened ability to use such cues. Previous studies have demonstrated the importance of temporal envelope cues in periodicity perception (e.g, gender recognition) by cochlear-implant (CI) listeners. The results of the present study suggest that that aging affects the use of such cues, and consequently gender recognition might be poorer among older CI recipients.

Keywords: aging, cochlear implants, pitch

Introduction

Voice-gender identification and discrimination play a significant role in everyday communication and multi-talker listening environments. Normal-hearing (NH) individuals are able to utilize voice-pitch cues to separate sound sources and extract a target talker from other distracting voices (Arehart et al., 1997; Summers & Leek, 1998; Brungart, 2001). Although multiple factors contribute to the perception of voice gender, the fundamental frequency (F0) of the talker commonly prevails as the most important cue (Murry & Singh, 1980; Klatt & Klatt, 1990). Researchers have debated the importance of spectral and temporal cues to pitch perception (Goldstein, 1973; Licklider, 1951; Meddis & Hewitt, 1991a; 1991b; Schouten, 1940; Terhardt, 1972; von Hemholtz, 1877; Wightman, 1973). While it is beyond the scope of this paper to review such theories, it has been repeatedly demonstrated that pitch perception is strongest when the lower harmonics are resolved (Ritsma, 1967; Shackleton & Carlyon, 1994; Qin & Oxenham, 2005); this ability, of course, requires adequate spectral resolution. However, these studies also demonstrate that the pitch of a harmonic complex can also be roughly perceived when the lower harmonics are removed (or unresolved) by utilizing periodicity information conveyed in the temporal envelope.

It has been repeatedly demonstrated that those with reduced spectral resolution, such as cochlear-implant (CI) users, demonstrate poor gender discrimination abilities (Fu et al., 2004; 2005; Gonzalez & Oliver, 2005; Vongphoe & Zeng, 2005). This is presumably due to the absence of temporal fine structure information and broadened auditory filters in CI users (primarily attributed to electrical current overlap). It is also important to note that information about talker-gender is available using formant-structure cues (Childers & Wu, 1991) via spectral peaks. Studies have shown, however, that spectral-peak resolution is poor among CI users (Henry & Turner, 2003; Henry, Turner, & Behrens, 2005). Recent evidence obtained among children suggests that, while children with CIs can perceive differences in voice pitch, spectral envelope cues do not strongly promote differentiation of talker-gender in CI users within this age group (Kovacic & Balaban, 2009; Vongpaisal et al,, 2010). In fact, overall evidence suggests that CI users rely primarily on temporal envelope cues to discern voice pitch (Chatterjee & Peng, 2008; Fu et al., 2004, 2005; Kovacic & Balaban, 2009).

Although temporal envelope periodicity cues result in weak F0 perception, several studies have demonstrated that CI users are able to discern differences in talker gender (primarily based on differences in F0 information). In such studies, performance was measured among CI listeners or among NH listeners using CI simulation techniques such as noiseband or sinewave vocoding. These methods mimic CI processing by systematically limiting the number of spectral channels and temporal envelope information available to the listener. Briefly, the original signal is divided into a given number of frequency bands then replaced with noisebands or sinewaves comprised of frequencies that approximate the original bandwidth. Each frequency band is then modulated with its respective temporal envelope, which can be manipulated (i.e., low-pass filtered) or left unaltered. The frequency bands are then summed to create the final vocoded signal (Shannon et al., 1995). Using this method, the number of channels (spectral cue) and low-pass cut-off frequency for temporal envelope extraction (temporal cue) can be systematically manipulated.

For example, Fu et al. (2004) measured gender identification in six younger NH adults (ages 22-30) when listening to sinewave vocoded speech and in 11 adult CI users (ages 39-70). Voice gender identification was measured using a closed-set, two-alternative forced-choice (2AFC) task. Stimuli consisted of 12 vowels varying in talker gender. For NH participants, the degree of spectral resolution was varied by providing 4-32 spectral channels, and the upper-limit of the temporal envelope was manipulated as well (20-320 Hz), thereby allowing access to varying degrees of temporal envelope cues. Results showed that performance was near perfect with 32 channels, but listeners exhibited systematic declines in performance as the number of channels was reduced. Further, performance improved as the temporal envelope cut-off frequency was increased (from 20 to 320 Hz) particularly when spectral cues were severely degraded thereby exhibiting a significant interaction between spectral and temporal cues. Results from subsequent, related studies corroborated those obtained by Fu et al., (2004) (Fu et al., 2005; Gonzolez & Olivier, 2005), and Fu et al. (2005) further demonstrated that the ability to utilize residual temporal envelope cues is reduced when F0 values are more similar across talkers. These results agreed with those obtained by Vongphoe & Zeng (2005) who found that CI listeners (and NH listening to CI simulations) experienced difficulty utilizing temporal envelope cues when required to discriminate between small differences in F0 in performing a talker identification task.

One important detail regarding the above studies (Fu et al., 2004; 2005; Vongphoe & Zeng, 2005) is the fact that a sinewave vocoding method was employed (instead of a noiseband vocoding method) the former of which often results in better performance on several tasks (Dorman et al., 1997; Gonzalez & Oliver, 2005) particularly when task performance was dependent upon F0 coding. Gonzalez & Oliver (2005) measured gender identification in 15 NH listeners (ages 21-30), when listening to either sinewave or noiseband vocoded stimuli. The original stimulus was a Spanish sentence, recorded by 40 native speakers of Spanish (20 males, 20 females). All stimuli were manipulated to contain 3- 16 channels, using sinewave and noiseband vocoding methods. The cut-off frequency of the temporal envelope was maintained at 400 Hz for sinewave carriers, and 160 Hz for noiseband carriers. Overall, results showed far better performance with the sinewave vocoded stimuli compared to the noiseband vocoded stimuli, particularly when the number of spectral channels was significantly reduced. The differences in performance can be readily explained, as a modulation of a sinewave generates side bands which convey information about the spectral content of the signal. For this reason, noiseband vocoding may better represent CI processing. Taken together, these studies show that gender and talker identification are difficult tasks among younger listeners when spectral cues are severely limited.

Despite such findings in younger listeners, the influence of spectral degradation on gender identification among older listeners has yet to be investigated. While some studies showed that speech comprehension may be affected by aging when stimuli are spectrally degraded (Peelle & Wingfield, 2005; Schvartz et al., 2008; Sheldon et al., 2008a; 2008b), there is also sufficient reason to speculate that aging affects spectrally-degraded voice pitch processing. For example, Vongpaisal and Pichora-Fuller (2007) showed that periodicity coding of resolved harmonics was slightly worse among older listeners compared to younger listeners when individuals were measured on an F0 discrimination task. Further studies demonstrate senescent decline in the perception of temporal envelope periodicity information. For example, Purcell and colleagues (Purcell et al., 2004) measured the envelope following responses in younger and older listeners, and demonstrated reduced amplitude in the older individuals but only for frequencies greater than 100 Hz. The authors hypothesized that such results support decreased temporal acuity in the aging auditory system, and in particular, a deficit in auditory brainstem function. These results are corroborated by subsequent electrophysiologic evidence (Leigh-Paffenroth & Fowler, 2006; Grose et al., 2009) and collectively endorse diminished periodicity coding in elderly listeners, but only at higher modulation rates (≥100 Hz). Lastly, neurophysiologic data also supported age-related differences in temporal envelope periodicity coding at the level of the inferior colliculus (Walton et al, 2002), particularly at higher modulation rates. Those modulations rates at which age-related differences in performance are most commonly observed (≥ 100 Hz) correspond to those which are important for voice-pitch information and talker gender discrimination.

In summary, it is widely believed that CI users utilize F0 information available in the temporal envelope to determine voice-pitch, and consequently talker-gender. While several studies have measured the ability of younger listeners to identify talker-gender when stimuli are spectrally degraded, it is possible that aging adversely affects this ability. Specifically, some studies provide support for age-related declines in temporal envelope periodicity processing (Leigh-Paffenroth & Fowler, 2006; Grose et al., 2009; Purcell et al., 2004; Walton et al., 2002). It is unknown, however, if these age-related differences obtained using strictly controlled, non-speech stimuli would translate to F0-based gender discrimination performance using a natural speech stimulus. The objective of the present investigation was to measure F0-based gender discrimination in younger and older NH listeners, under conditions of spectral and temporal degradation. It was hypothesized that older listeners would have greater difficulty under conditions with a reduced number of spectral channels due to a greater reliance on temporal envelope cues.

Methods and Materials

Participants

Ten younger (ages 21-28) and ten older (ages 60-73) male and female NH listeners participated in the current investigation. For the purposes of this study, NH was defined as air-conduction audiometric thresholds ≤ 20 dB HL from 250-4000 Hz in the test ear (ANSI, 2004). Average audiometric thresholds and corresponding standard deviations for each group are provided in Table I. These values reflect data obtained in the test ear only. All participants reported to be in good health and denied a significant history of audiologic or otologic disease.

Table I.

Average audiometric pure-tone thresholds for each frequency are provided for each age group (Mean), with the corresponding standard deviation provided below (SD). The pure-tone average (PTA) is also provided. The reported values represent data from participants’ test ear. The asterisk refers to frequencies (6000 and 8000 Hz) that were tested, however were not used to determine inclusion into the study, as the experimental stimuli contained information up to 4000 Hz only.

Audiometric data for each group
PTA 250 500 1000 2000 3000 4000 6000* 8000*
YNH Mean 6.33 9.5 7 6.5 5.5 7.5 3.5 7.5 6.5
SD 3.58 5.98 4.83 4.11 5.98 5.97 4.74 4.26 5.29
ONH Mean 10.5 15 8.5 12 11 13.5 16.5 19.5 27
SD 1.93 5.27 4.11 4.21 3.16 3.37 4.11 8.31 16.36
Open in a new tab

Stimuli

Stimuli were six naturally produced and digitally recorded (sampling rate =22,050 Hz) vowel tokens (/h/vowel/d/ context, Hillenbrand et al., 1995): /æ/, /i/, /oInline graphic/, /Inline graphic/, /eInline graphic/, /u/. Each of the six recorded vowels was spoken by four male and four female talkers, for a total of 48 vowels (6 vowels × 8 talkers). Vowels were selected from a larger corpus consisting of 144 vowels, originally recorded at the House Ear Institute (HEI) and were available within the Computer-Assisted-Speech-Training (CAST) software developed by Qian-Jie Fu at HEI (Fu, 2008).

The F0 values were analyzed and manipulated using the autocorrelation method (Boersma, 1993) available in Praat (version 5.0.27) (Boersma & Weenink, 2008). Specifically, the average F0 value of each token was calculated then shifted to match a specified target (within 1 Hz) for a given condition and talker. Assignment of target F0 values was dependent upon the original F0 value of each token; that is, the lowest F0 value among the original tokens was assigned the lowest target F0 value. These F0 targets were selected to create two general sets of stimuli for the current investigation: ‘Expanded’ and ‘Compressed’. For the Expanded condition, F0 values ranged from 100 Hz (lowest F0 of male voice) to 275 Hz (highest F0 value of female voice). For the Compressed condition, the range of F0 values was reduced so that the lowest value was 150 Hz (lowest F0 of male voice) to 220 Hz (highest F0 value of female voice). Each manipulated token was stored on computer hard drive then uploaded into the CAST program at the time of testing (see below). Table II and Table III provide the target average F0 values (and F0 range) for each talker in the Expanded and Compressed conditions, respectively. Based on these values, average difference in F0s between the male and female talkers was 100 Hz and 40 Hz for the Expanded and Compressed conditions, respectively. It is important to point out that while the average pitch was manipulated, the natural pitch fluctuations within a token remained unaltered.

Table II.

Target average fundamental frequency (F0) values and actual ranges (Hz) estimated for each talker, for the Expanded condition. Actual average F0 values for each vowel stimulus may have varied by ±1 Hz for each talker.

F0 values (Expanded)
F0 hInline graphicd hInline graphicd hInline graphicd h/eInline graphicd hInline graphicd hInline graphicd
Min Max Min Max Min Max Min Max Min Max Min Max
100 98 102 97 105 96 104 95 106 98 103 98 104
125 120 138 123 128 122 128 121 128 122 128 124 127
150 144 155 146 152 147 154 145 154 144 156 144 153
175 169 185 170 182 170 185 170 189 171 178 172 188
200 190 222 182 225 182 227 183 232 180 238 185 234
225 209 243 219 230 220 235 213 245 212 248 210 241
250 240 268 242 262 245 265 242 263 243 261 241 265
275 237 290 238 286 264 300 260 285 269 283 270 290
Open in a new tab

Table III.

Identical to Table II, but representative of stimuli used in the Compressed condition.

F0 values (Compressed )
F0 hInline graphicd hInline graphicd hInline graphicd h/eInline graphicd hInline graphicd hInline graphicd
Min Max Min Max Min Max Min Max Min Max Min Max
150 148 151 147 154 144 154 144 157 147 152 148 153
160 154 164 157 162 157 164 157 162 156 163 159 161
170 165 176 167 172 165 174 164 173 164 178 163 174
180 175 190 175 187 175 193 178 192 176 185 177 194
190 180 224 171 217 170 220 171 224 169 230 175 226
200 182 221 192 204 195 210 187 226 186 224 183 220
210 200 229 201 222 204 225 201 224 203 221 201 226
220 191 236 182 241 175 255 160 256 172 250 163 247
Open in a new tab

Noiseband vocoding was accomplished online using TigerCIS within the CAST interface. This was done after each of the pitch manipulations described above. Noiseband vocoding methods used in the current study were comparable to those described in Shannon et al., (1995). Stimuli were first band-pass filtered into 1, 4, 8, 16, or 32 channels depending on the specific condition, using fourth-order Butterworth filters (24 dB/octave).

All stimuli were filtered from 100-4000 Hz, and specific division frequencies for each case (i.e. number of channels) were determined based on the logarithmic equation provided by Greenwood (1990). See Table IV for division frequency values. The temporal envelope was then extracted from each frequency band using half-wave rectification and low-pass filtering. The cut-off frequency of the low-pass filter was 20, 50, 100, 200 or 400 Hz, depending on the condition. The original fine structure of each filter was then replaced with corresponding band-pass noise (fourth-order Butterworth filters), and the temporal envelope was used to modulate the noise bands. The modulated noise bands were summed and then normalized to a specified but equal RMS value to result in the final noiseband vocoded speech token.

Table IV.

Division frequencies used for the noise-band vocoding processing. The division number is listed in the first column, with the corresponding condition (number of channels) listed towards the top of the table.

Division frequencies for noise-band vocoding filters
(Hz)
Number of channels
Division
#		 32 16 8 4
1 123 147 204 352
2 147 204 352 863
3 174 272 563 1900
4 204 352 863
5 236 448 1291
6 272 563 1900
7 310 700 2766
8 352 863
9 398 1058
10 448 1291
11 503 1568
12 563 1900
13 629 2295
14 700 2766
15 778 3329
16 863
17 957
18 1058
19 1169
20 1291
21 1424
22 1568
23 1727
24 1900
25 2088
26 2295
27 2520
28 2766
29 3035
30 3329
31 3650
Open in a new tab

In order to demonstrate how noise band vocoding manipulates spectral and temporal cues, amplitude spectra and time-amplitude waveforms of the vowel h/Inline graphic/d are shown in Figures 1 and 2, respectively, with each figure containing varying degrees of spectral resolution (number of channels). In each case the vowel was spoken by the same male talker (F0= 100 Hz), and the cut-off frequency of the low-pass temporal envelope filter is equal to 400 Hz in all cases. As shown in Figure 1, spectral peaks become obscured with fewer spectral channels. Likewise, as shown in Figure 2, the temporal fine structure information is degraded when stimuli are noiseband vocoded, but the temporal envelope periodicity information is available even when spectral cues are absent. In this case, the envelope modulates at the rate of the F0, 100 Hz.

Figure 1.

Figure 1

Open in a new tab

Amplitude spectra for the vowel h/Inline graphic/d spoken by a male with an average F0=100 Hz. As indicated in the legend, each line represents varying degrees of spectral resolution (number of spectral channels).

Figure 2.

Figure 2

Open in a new tab

Time-amplitude waveforms for the vowel h/Inline graphic/d spoken by a male with an average F0=100 Hz. Each panel (2A-2D) represents the same token processed with varying degrees of spectral cues (A= Unprocessed, B= 16 channels, C= 8 channels, D = 1 channel). The cut-off frequency of the low-pass temporal envelope filter is equal to 400 Hz in panels B, C and D. The small insert within each panel represents a close examination of the temporal fine structure within each waveform.

The channel and envelope conditions differed in the Expanded and Compressed stimuli; this was decided based on pilot data that showed the Expanded conditions to be easier than the Compressed conditions. For example, performance reached nearly 100% with 16 channels in the Expanded condition therefore the 32 channel condition was omitted from the test protocol. Channel and envelope conditions were adjusted accordingly within the Expanded and Compressed conditions, thereby resulting in four channel/five envelope conditions for Expanded stimuli, and five channel/four envelope conditions for Compressed stimuli.

Procedure

A two-alternative forced-choice (2AFC) method was used to measure gender identification abilities through the CAST graphical user interface. Using a computer mouse and screen, the participant was instructed to select “start” in order to hear the first vowel token. Following the presentation, two rectangular boxes appeared on the computer screen: one labeled “Male” and the other “Female”. The subject was instructed to identify the gender of the speaker by selecting the appropriate box; there was no time limit for responding. After making a selection the next stimulus was presented 1 second later. This continued until the end of a run, which consisted of 48 vowel tokens (1 repetition × 8 talkers × 6 vowels).

In order to reduce the possibility of order effects, half of the listeners were first presented with stimuli from the Expanded condition, whereas half of the listeners were first presented with stimuli from the Compressed condition. Prior to being tested using the degraded stimuli for each condition, listeners listened to one run of unprocessed stimuli; all listeners were required to achieve 90% accuracy on this task before continuing on in the study. No subject was excluded from the study based on this criterion. Participants were then presented with three presentation cycles, each consisting of a random ordering of all possible noiseband vocoded conditions. Specifically, a cycle consisted of 40 runs (20 Expanded − 4 channel conditions × 5 envelope conditions plus 20 Compressed − 5 channel conditions × 4 envelope conditions). Within each of the 40 runs, the 48 vowel tokes were presented in a completely random order. The first presentation cycle served as practice, and an average percent correct was calculated based on the final two cycles.

Stimuli were output through an external soundcard (Edirol 25-UAEX) and mixer (Rane SM26B), before being delivered monaurally through a calibrated circumaural headphone (Sennheiser HDA 200) at a level of 65 dB SPL. For most listeners, stimuli were presented to the right ear with the exception of those listeners (N=2) whose right-ear audiometric thresholds did not meet the criteria for normal hearing (as defined under “Participants” above). In this case, stimuli were presented to the better-hearing (left) ear. Testing was performed in two-hour time blocks, with breaks provided at the listener’s discretion. The total duration of the experiment varied from 6-10 hours (3-5 sessions).

Results

Prior to all analyses, percent correct scores were converted to rationalized arsine units (RAUs) (Studebaker, 1985). If it was found that the assumption of sphericity was violated when conducting an analysis of variance (ANOVA), the Greenhouse-Geisser correction was used to interpret the results.

Overall effects of age on gender identification

Results from the Expanded condition are shown in Figure 3, while results from the Compressed condition are shown in Figure 4. A mixed ANOVA was performed within the Expanded and Compressed conditions, to quantify the interactions between age group, as well as availability of spectral and temporal cues. Subject group (age) served as the between-group factor, whereas number of spectral channels (channels) and cut-off frequency of the low-pass temporal envelope filter (envelope) served as the within-group factors.

Figure 3.

Figure 3

Open in a new tab

Results are shown for performance on the gender identification task for the Expanded condition. Each graph represents results obtained within each group (panel ‘A’ = younger listeners and panel ‘B’ = older listeners) for each condition, with the number of channels represented along the abscissa and percent correct along the ordinate. The label “ALL” refers to performance using unprocessed stimuli. The various symbols represent the value of the cut-off frequency of the low-pass temporal envelope filter (see legend). The error bars represent ±1 standard deviation from the mean. Chance performance (50%) is indicated by the dashed horizontal line within each graph.

Figure 4.

Figure 4

Open in a new tab

Identical to Figure 3, but represents data obtained in the Compressed conditions.

As expected, performance was better for Expanded than the Compressed condition. The between-group differences, however, varied within each condition. Overall results suggest that, consistent with previous findings, performance improved as the number of spectral channels and temporal envelope cut-off frequency increased. This general pattern was true for both age groups. Also consistent with previous research (Fu et al., 2004; 2005), results demonstrated a significant interaction between the use of spectral and temporal cues: as the number of spectral channels was decreased, listeners showed greater benefit from increasing temporal envelope information. Perhaps, most importantly the results suggested poorer voice gender discrimination among older compared to younger listeners.

For the Expanded condition, there was a significant main effect of channels [F (3,54) = 441.84, p<0.01] , a significant main effect of envelope [F (2.10, 37.91) = 33.99, p<0.01], and a significant interaction between channel and envelope [F (5.13, 92.34)=5.33, p<0.01]. Further, there was a significant main effect of age [F (1,18) = 16.89, p<0.01], as the average performance of the younger NH group exceeded that of the older NH group. Follow-up testing (using multiple paired t-tests with Bonferroni corrections) was performed for the significant channel × envelope interaction. Overall, the results from this analysis showed that there was trade-off between spectral and temporal cues. Specifically, listeners’ showed greater benefit from increasing temporal envelope information (i.e., increasing low-pass filter cut-off frequency) as the number of spectral channels was reduced. For example, for the 1-channel condition, performance significantly improved by 16.3 RAUs, on average, from 20 to 400 Hz (t= −5.94 , p< 0.001), whereas there was no effect of temporal envelope cut-off frequency in the 16-channel condition.

The ANOVA results for the Compressed condition show that there was a significant main effect of channels [F (2.63,72) = 579.06, p<0.01), a significant main effect of envelope [F (3,54) = 10.63, p<0.05], and a significant main effect of age [F (1, 18) = 7.55, p<0.05]. There was also a significant age × envelope interaction [F (3,54) = 3.40, p<0.05). Unlike results for the Expanded condition, the interaction between channels and envelope was not significant. Follow-up tests (one-way ANOVA followed by paired t-test with Bonferroni correction) were used to further examine the interaction between age and envelope, and results revealed that was a significant main effect of envelope within the YNH group [F (3,147) = 13.9, p<0.01], but not within the older NH group. More specifically, the performance of the younger NH group improved by 3.8 – 5.8 RAUs, on average, as the temporal envelope cut-off frequency was increased from 50 to 200 Hz, from 50 to 400 Hz, and from 100-200 Hz (p<0.001). Although these are not large increases in performance, they are significant, which is certainly meaningful given the similarity of F0 values in the Compressed condition. Lastly, it is important to note that a two-way ANOVA showed no significant differences between expanded and compressed conditions or between age groups when listeners performed the gender identification task.

Talker analyses

In order to further analyze the effects of aging on gender identification, we examined gender identification performance with respect to each of the eight talkers (four male, four female). We expanded the mixed ANOVA analyses (described in the previous section) to include the additional within-group factor of talker-F0. We hypothesized that, the between group differences noted in the initial analyses are dependent upon the individual talkers. It is hypothesized that both age groups were able to better identify talker gender when the F0 of the talker is more extreme (lower F0 for a male, higher F0 for a female). However, we anticipated that the improvement in performance as talker-F0 became more extreme would be less in the older compared to the younger groups.

Results from the Expanded condition are shown in Figure 5, while results from the Compressed condition are shown in Figure 6. Within each of these figures, data plotted in the left and right panels represent results obtained in the younger and older groups, respectively. Within each panel, talker-F0 is shown along the abscissa, while performance on the gender identification task is shown along the ordinate (in percent correct). Analyses within the Expanded condition showed a significant interaction between talker and channel [F (3.5, 63.5) = 6.35, p < 0.01], and follow-up testing (t-tests with Bonferroni correction) indicated that while performance did tend to improve as the talker-F0 became lower (male) or higher (female) it was also dependent on the number of channels available to the listener. For example, in the 1-channel condition performance systematically improved as the F0 of a male talker decreased (at each 25 Hz decrease from 175 to 100 Hz, p<0.05). Further, the same pattern does not hold true as the number of spectral channels increases and spectral cues become more available: as shown in Figure 5, performance is fairly equivalent regardless of talker-F0 in the 16 channel condition. It is possible, however, that these results could be subject to ceiling effects. Results also indicated a significant interaction between talker and age group [ F(7, 126) = 2.407, p<0.05] ;this overall finding was not dependent on the number of channels (no significant interaction between talker-F0, age group, and number of channels).

Figure 5.

Figure 5

Open in a new tab

Gender identification performance is shown according to each individual talker for the Expanded condition. The panels on the left and right show data from the younger and older listeners, respectively. The top panels show data obtained when stimuli were processed to contain 1-channel, and the lower panels show data obtained as more spectral channels were made available to the listener. Within each graph, the thin lines (open symbols) represent varying degrees of temporal envelope cues (see legend), while the thick line (closed symbol) represents average performance collapsed across all temporal envelope conditions. The horizontal dashed line represents chance performance (50%). Error bars represent ±SE from the mean.

Figure 6.

Figure 6

Open in a new tab

Identical to Figure 5, but represents data obtained in the Compressed conditions.

Analyses for the Compressed condition revealed performance patterns similar to those for the Expanded condition. However, in this case the interaction between talker-F0 and age group was not significant. While Figure 6 shows that performance was dependent on talker-F0, this effect did not depend on the age of the listener. This result would suggest that younger listeners are better able to use differences in talker-F0 to improve gender identification performance but only when these values are more diverse.

Taken together, these results are somewhat consistent with our hypothesis that, in some cases, older listeners are not able to use temporal envelope F0 cues to differentiate talker gender to the same extent as their younger counterparts. Specifically, analyses suggest that when F0 values of talkers are largely different from one another (as in the Expanded conditions), younger listeners are better able to use temporal envelope cues to determine if the stimulus was spoken by a male or female compared to older listeners. This effect is clearly shown in Figure 5: while the overall performance of older listeners is poorer than that of the younger listeners, the two age groups exhibit different patterns of performance dependent upon the temporal envelope information. When spectral cues are reduced (Figure 5, 1 channel condition), the performance of younger listeners improves with increasing temporal envelope information (for both male and female talkers). The performance of the older listeners, however, is less dependent on temporal envelope information (particularly for female voices/higher F0 values). This pattern may suggest that, in some cases, older listeners cannot use higher rate temporal envelope information to the same extent as their younger counterparts, even when such cues are made readily available to the listener.

Discussion

In keeping with previous investigations (Fu et al., 2004; 2005), the listeners who participated in the present study demonstrated a systematic relationship between the utilization of temporal and spectral cues when listening in a quiet background. Spectral cues undoubtedly provide better resolution when coding voice-pitch information, and temporal envelope cues are seemingly extraneous when listeners are provided with 16 or 32 spectral channels. However, when spectral cues are absent, some listeners are able to utilize temporal envelope cues to perceive voice-pitch, albeit to a lesser extent than when all spectral cues are available.

The results presented in the current study expand upon those reported in previous investigations measuring temporal envelope processing in aging adults. Previous neurophysiological, electrophysiological, and psychoacoustic studies have alluded to the presence of senescent decline in envelope periodicity processing, while the current study demonstrated that such temporal processing deficits translate to voice pitch coding of natural speech when spectral cues are decreased or absent altogether. Overall, it appears that, while older listeners were able to utilize temporal envelope cues to perceive voice pitch in some cases, the resolution and discrimination of temporal envelope modulation rate was worse than that of their younger NH peers. For example, in the current study we implemented two main conditions: one in which the range of F0 values was large (Expanded), and the second in which the range of F0 values was compressed (Compressed). Based on the results of the Expanded condition, older listeners were able to sometimes discriminate voice gender, even when spectral cues are severely degraded as long as sufficient temporal envelope cues were available. In doing so, they were presumably coding voice pitch using differences in the modulation rates of the residual temporal envelope (see Figure 2D). Results from the Compressed condition, however, revealed that older listeners lack the resolution to discriminate between residual temporal envelope modulation rates when F0 values (and consequently amplitude modulation rates) are more similar to one another. These findings agree with previous reports of impaired periodicity coding in older listeners for high rate information (Leigh-Paffenroth & Fowler, 2006; Grose et al., 2009; Pichora-Fuller et al., 2007; Purcell et al., 2004; Walton et al., 2002; Walton, 2010).

Although several cues contribute to the perception of talker-gender, further analyses indicated that listeners in the current study were indeed using F0 cues in order to perform the gender identification task. This was particularly true as spectral cues became severely degraded, in which case performance was presumably driven by differences in the modulation rate of the temporal envelope.

Collectively, evidence suggests that CI listeners are able to utilize 8-10 spectral channels, at best (Friesen et al., 2001). Looking at the results for the 8-channel conditions in the current study, it appears that even younger CI users would experience difficulty coding voice-pitch information, as the younger NH participants’ performance was 92.5 and 79.4 RAUs for the Expressed and Compressed conditions, respectively when all temporal envelope cues were available (i.e.,400 Hz cut-off). Older listeners, however, exhibited poorer performance in the 8-channel condition when compared to their younger NH peers, as they achieved average scores of 80.9 and 74.1 RAUs for Expanded and Compressed conditions, respectively (i.e.,400 Hz cut-off).

Although little is known about the subtle but potentially significant interactions between aging and cochlear implantation, CIs are undoubtedly beneficial for older listeners (Vermeire et al., 2005; Chatelin et al., 2004). Based on the results of this study (along with a growing number of publications citing temporal envelope processing deficits in older listeners), it is reasonable to conjecture that older CI recipients might be at a disadvantage when distinguishing between talker genders. One caveat to this conclusion is the fact that electrical stimulation of the auditory nerve results in greater sensitivity to temporal amplitude modulations, compared to acoustic stimulation (Shannon, 1992). Another point of debate is the extent to which using CI simulations reflects the performance of older CI users, particularly when relying on temporal cues. Specifically, older adults with hearing loss (CI recipients) may exhibit even greater temporal processing deficits compared to older adults with normal peripheral hearing (Lorenzi et al., 2009). Therefore, it is possible that results of the current study underestimate temporal processing deficits of CI recipients. Lastly, there is little doubt that training and listening experience would improve CI users’ performance, but it impossible to predict how aging would influence performance over time. Even though the results presented in the current study do not reveal an extraordinary effect of age on the spectro-temporal processing of voice-pitch cues, the implication for older CI users could be rather considerable given the importance of F0 coding in daily communication (Arehart et al., 1997; Summers & Leek, 1998; Brungart, 2001).

Conclusion

In summary, the current investigation demonstrated a significant interaction between spectro-temporal envelope processing and the aging auditory system. The results of this study complement those of previous investigations that cited age-related changes in temporal periodicity processing when non-speech stimuli were used. The present findings suggest that such age-related temporal deficits translate to changes in voice-pitch processing of spectrally degraded speech stimuli. Collectively, these findings could have important implications for the rehabilitation of older adult CI users.

Acknowledgements

Many thanks to Qin-Jie Fu, Ph.D. (House Ear Institute) for software support and use of his research interface. We would like to thank the subjects for their time and willingness to participate in the current study. This work was funded by NIH/NIDCD grant no. R01DC004786 to MC, and training grant T32DC000046 to KCS.

Funding acknowledgments: This work was funded by NIH/NIDCD grant no. R01DC004786 to MC, and training grant T32DC000046 to KCS.

Footnotes

1

Portions of this work were presented at The 2009 Meeting of the American Auditory Society, Scottsdale, AZ.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

References

  1. ANSI . American National Standard Specification for Audiometers. American National Standards Institute; New York: 2004. ANSI S3.6-2004. [Google Scholar]
  2. Arehart KH, King CA, McLean-Mudgett KS. Role of fundamental frequency differences in the perceptual separation of competing vowel sounds by listeners with normal hearing and listeners with hearing loss. J. Speech Hear Lang. Res. 1997;40:1434–1444. doi: 10.1044/jslhr.4006.1434. [DOI] [PubMed] [Google Scholar]
  3. Boersma P. Proceedings of the Institute of Phonetic Sciences. Vol. 17. University of Amsterdam; 1993. Accurate short-term analysis of the fundamental frequency and the harmonic-to-noise ratio of a sampled sound; pp. 97–110. [Google Scholar]
  4. Boersma P, Weenink D. [Retrieved: October 1, 2008];Praat: doing phonetics by computer. (Version 5.0.34). 2008 [Computer Program] from http://www.praat.org.
  5. Brungart DS. Informational and energetic masking effects in the perception of two simultaneous talkers. J. Acoust. Soc. Am. 2001;109:1101–1109. doi: 10.1121/1.1345696. [DOI] [PubMed] [Google Scholar]
  6. Chatelin V, Kim EJ, Driscoll C, Larky J, Polite C, Price L, Lalwani A. Cochlear implant outcomes in the elderly. Otol Neurotol. 2004;25:298–301. doi: 10.1097/00129492-200405000-00017. [DOI] [PubMed] [Google Scholar]
  7. Chatterjee M. Modulation masking in cochlear implant listeners: envelope versus tonotopic components. J. Acoust. Soc. Am. 2003;113:2042–2053. doi: 10.1121/1.1555613. [DOI] [PubMed] [Google Scholar]
  8. Chatterjee M, Oba SI. Across- and within-channel envelope interactions in cochlear implant listeners. J. Assoc. Res. Otolaryngol. 2004;5:360–375. doi: 10.1007/s10162-004-4050-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Childers DG, Wu K. Gender recognition from speech. Part II: Fine analysis. J. Acoust. Soc. Am. 1991;90:1841–1856. doi: 10.1121/1.401664. [DOI] [PubMed] [Google Scholar]
  10. Dorman MF, Loizou PC, Rainey D. Speech intelligibility as a function of the number of channels of stimulation for signal processors using sine-wave and noise-band outputs. J. Acoust. Soc. Am. 1997;102:2403–2411. doi: 10.1121/1.419603. [DOI] [PubMed] [Google Scholar]
  11. Friesen LM, Shannon RV, Baskent D, Wang X. Speech recognition in noise as the number of spectral channels: Comparison of acoustic hearing and cochlear implants. J. Acoust. Soc. Am. 2001;110:1150–1163. doi: 10.1121/1.1381538. [DOI] [PubMed] [Google Scholar]
  12. Fu Q-J, Chinchilla S, Galvin JJ. The role of spectral and temporal cues in voice gender discrimination by normal-hearing listeners and cochlear implant users. J. Assoc. Res. Otolaryngol. 2004;5:253–260. doi: 10.1007/s10162-004-4046-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Fu Q-J, Chinchilla S, Nogaki G, Galvin JJ. Voice gender identification by cochlear implant users: The role of spectral and temporal resolution. J. Acoust. Soc. Am. 2005;118:1711–1718. doi: 10.1121/1.1985024. [DOI] [PubMed] [Google Scholar]
  14. Fu Q-J. [Retrieved: 2008];TigerSpeech Technology: Computer Assisted Speech Training. (Version 5.04.02). 2008 [Computer Program] from www.tigerspeech.com.
  15. Goldstein JL. An optimum processor theory for the central formation of the pitch of complex tones. Journal of the Acoustical Society of America. 1973;54:1496–1516. doi: 10.1121/1.1914448. [DOI] [PubMed] [Google Scholar]
  16. Gonzalez J, Oliver JC. Gender and speaker identification as a function of the number of channels in spectrally reduced speech. J. Acoust. Soc. Am. 2005;118:461–470. doi: 10.1121/1.1928892. [DOI] [PubMed] [Google Scholar]
  17. Greenwood DD. A cochlear frequency-position function for several species—29 years later. J. Acoust. Soc. Am. 1990;87:2592–2605. doi: 10.1121/1.399052. [DOI] [PubMed] [Google Scholar]
  18. Grose JH, Mamo SK, Hall JW. Age effects in temporal envelope processing: Speech unmasking and auditory steady state responses. Ear Hear. 2009;30:568–575. doi: 10.1097/AUD.0b013e3181ac128f. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Henry BA, Turner CW. The resolution of complex spectral patterns by cochlear implant and normal hearing listeners. J. Acoust. Soc. Am. 2003;113:2861–2873. doi: 10.1121/1.1561900. [DOI] [PubMed] [Google Scholar]
  20. Henry BA, Turner CW, Behrens A. Spectral peak resolution and speech recognition in quiet: normal hearing, hearing impaired, and cochlear implant listeners. J. Acoust. Soc. Am. 2005;118:1111–1121. doi: 10.1121/1.1944567. [DOI] [PubMed] [Google Scholar]
  21. Hillenbrand J, Getty LA, Clark MJ, Wheeler K. Acoustic characteristics of American Englishvowels. J. Acoust. Soc. Am. 1995;97:3099–3111. doi: 10.1121/1.411872. [DOI] [PubMed] [Google Scholar]
  22. Klatt DH, Klatt LC. Analysis, synthesis, and perception of voice quality variations among female and male talkers. J. Acoust. Soc. Am. 1990;87:820–857. doi: 10.1121/1.398894. [DOI] [PubMed] [Google Scholar]
  23. Kovacic D, Balaban E. Voice gender perception by cochlear implantees. J. Acoust. Soc. Am. 2009;126:762–775. doi: 10.1121/1.3158855. [DOI] [PubMed] [Google Scholar]
  24. Leigh-Paffenroth ED, Fowler CG. Amplitude-modulated auditory steadystate responses in younger and older listeners. J. Am. Acad. Aud. 2006;17:582–597. doi: 10.3766/jaaa.17.8.5. [DOI] [PubMed] [Google Scholar]
  25. Licklider JCR. A duplex theory of pitch perception (reproduced in Schubert, 1979, 155-160) Experientia. 1951;7:128–134. doi: 10.1007/BF02156143. [DOI] [PubMed] [Google Scholar]
  26. Lorenzi C, Debruille L, Garnier S, Fleuriot P, Moore BC. Abnormal processing of temporal fine structure in speech for frequencies where absolute thresholds are normal. J. Acoust. Soc. Am. 2009;125:27–30. doi: 10.1121/1.2939125. [DOI] [PubMed] [Google Scholar]
  27. Meddis R, Hewitt MJ. Virtual pitch and phase sensitivity of a computer model of the auditory periphery. I: Pitch identification. Journal of the Acoustical Society of America. 1991a;89:2883–2894. [Google Scholar]
  28. Meddis R, Hewitt MJ. Virtual pitch and phase sensitivity of a computer model of the auditory periphery. II: Phase sensitivity. J. Acoust. Soc. Am. 1991b;91:233–245. [Google Scholar]
  29. Murry T, Singh S. Multidimentional analysis of male and female voices. J. Acoust. Soc. Am. 1980;68:1294–1300. doi: 10.1121/1.385122. [DOI] [PubMed] [Google Scholar]
  30. Peelle JE, Wingfield A. Dissociations in perceptual learning revealed by adult age differences in adaptation to time-compressed speech. J Exp Psychol Hum Percept Perform. 2005;31:1315–1330. doi: 10.1037/0096-1523.31.6.1315. [DOI] [PubMed] [Google Scholar]
  31. Pichora-Fuller MK, Schneider BA, MacDonald E, Pass HE, Brown S. Temporal jitter disrupts speech intelligibility: A simulation of auditory aging. Hear. Res. 2007;223:114–1121. doi: 10.1016/j.heares.2006.10.009. [DOI] [PubMed] [Google Scholar]
  32. Purcell DW, John SM, Schneider BA, Picton TW. Human temporal auditory acuity as assessed by envelope following responses. J. Acoust. Soc. Am. 2004;116:3581–3593. doi: 10.1121/1.1798354. [DOI] [PubMed] [Google Scholar]
  33. Qin MK, Oxenham AJ. Effects of envelope-vocoder processing on F0 discrimination and concurrent-vowel identification. Ear Hear. 2005;26:451–460. doi: 10.1097/01.aud.0000179689.79868.06. [DOI] [PubMed] [Google Scholar]
  34. Ritsma RJ. Frequencies dominant in the perception of the pitch of complex sounds. Journal of the Acoustical Society of America. 1967;42:191–198. doi: 10.1121/1.1910550. [DOI] [PubMed] [Google Scholar]
  35. Schvartz KC, Chatterjee M, Gordon-Salant S. Recognition of spectrally degraded phonemes by younger, middle-aged, and older normal-hearing listeners. J Acoust Soc Am. 2008;124:3972–3988. doi: 10.1121/1.2997434. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Shackleton TM, Carlyon RP. The role of resolved and unresolved harmonics in pitch perception and frequency modulation discrimination. J. Acoust. Soc. Am. 1994;95:3529–3540. doi: 10.1121/1.409970. [DOI] [PubMed] [Google Scholar]
  37. Shaddock-Palombi P, Backoff PM, Caspary DM. Responses of young and aged rat inferior colliculus neurons to sinusoidally amplitude modulated stimuli. Hear Res. 2001;153:174–180. doi: 10.1016/s0378-5955(00)00264-1. [DOI] [PubMed] [Google Scholar]
  38. Shannon RV. Temporal modulation transfer functions in patients with cochlear implants. J. Acoust. Soc. Am. 1992;91:2156–2164. doi: 10.1121/1.403807. [DOI] [PubMed] [Google Scholar]
  39. Shannon RV, Zeng FG, Kamath V, Wygonski J, Ekilid M. Speech recognition with primarily temporal cues. Science. 1995;270:303–304. doi: 10.1126/science.270.5234.303. [DOI] [PubMed] [Google Scholar]
  40. Sheldon S, Pichora-Fuller MK, Schneider BA. Priming and sentence context support listening to noise-vocoded speech by younger and older adults. J Acoust Soc Am. 2008a;123:489–499. doi: 10.1121/1.2783762. [DOI] [PubMed] [Google Scholar]
  41. Sheldon S, Pichora-Fuller MK, Schneider BA. Effect of age, presentation method, and learning on identification of noise-vocoded words. J Acoust Soc Am. 2008b;123:476–488. doi: 10.1121/1.2805676. [DOI] [PubMed] [Google Scholar]
  42. Schouten JF. The residue and the mechanism of hearing. Proceedings of the Koninklijke Nederlandse Akademie van Wetenschappen, Series A. 1940;43:991–999. [Google Scholar]
  43. Studebaker GA. A ‘rationalized’ arcsine transform. J. Speech. Lang. Hear. Res. 1985;28:455–463. doi: 10.1044/jshr.2803.455. [DOI] [PubMed] [Google Scholar]
  44. Summers V, Leek MR. F0 processing and the separation of competing speech signals by listeners with normal hearing and with hearing loss. J. Speech. Lang. Hear. Res. 1998;41:1294–1306. doi: 10.1044/jslhr.4106.1294. [DOI] [PubMed] [Google Scholar]
  45. Terhardt E. Pitch, consonance and harmony. Journal of the Acoustical Society of America. 1974;55:1061–1069. doi: 10.1121/1.1914648. [DOI] [PubMed] [Google Scholar]
  46. Vermeire K, Brokx JPL, Wuyts FL, Cochet E, Hofkens A, Van de Heyning PH. Quality of life benefit from cochlear implantation in the elderly. Otol Neurotol. 2005;26:188–195. doi: 10.1097/00129492-200503000-00010. [DOI] [PubMed] [Google Scholar]
  47. von Helmholtz H. On the sensations of tone (English translation A.J. Ellis, 1885, reprinted 1954) Dover; New York: 1877. [Google Scholar]
  48. Vongpaisal T, Pichora-Fuller MK. Effect of age on F0 difference limen and concurrent vowel identification. J Speech Hear Lang Res. 2007;50:1139–1156. doi: 10.1044/1092-4388(2007/079). [DOI] [PubMed] [Google Scholar]
  49. Vongpaisal T, Trehub SE, Schellenberg EG, van Lieshout P, Papsin BC. Children with cochlear implants recognize their mother’s voice. Ear Hear. 2010;31:555–566. doi: 10.1097/AUD.0b013e3181daae5a. [DOI] [PubMed] [Google Scholar]
  50. Vongphoe M, Zeng FG. Speaker recognition with temporal cues in acoustic and electric hearing. J. Acoust. Soc. Am. 2005;118:1055–1061. doi: 10.1121/1.1944507. [DOI] [PubMed] [Google Scholar]
  51. Walton JP, Simon H, Frisina RD. Age-related alterations in the neural coding of envelope periodicities. J. Neurophysiol. 2002;88:565–578. doi: 10.1152/jn.2002.88.2.565. [DOI] [PubMed] [Google Scholar]
  52. Wightman FL. The pattern transformation model of pitch. Journal of the Acoustical Society of America. 1973;54:407–416. doi: 10.1121/1.1913592. [DOI] [PubMed] [Google Scholar]

RESOURCES