The search for correlates of age-related cochlear synaptopathy: Measures of temporal envelope processing and spatial release from speech-on-speech masking

Abstract

Older adults often experience difficulties understanding speech in adverse listening conditions. It has been suggested that for listeners with normal and near-normal audiograms, these difficulties may, at least in part, arise from age-related cochlear synaptopathy. The aim of this study was to assess if performance on auditory tasks relying on temporal envelope processing reveal age-related deficits consistent with those expected from cochlear synaptopathy. Listeners aged 20 to 66 years were tested using a series of psychophysical, electrophysiological, and speech-perception measures using stimulus configurations that promote coding by medium- and low-spontaneous-rate auditory-nerve fibers. Cognitive measures of executive function were obtained to control for age-related cognitive decline. Results from the different tests were not significantly correlated with each other despite a presumed reliance on common mechanisms involved in temporal envelope processing. Only gap-detection thresholds for a tone in noise and spatial release from speech-on-speech masking were significantly correlated with age. Increasing age was related to impaired cognitive executive function. Multivariate regression analyses showed that individual differences in hearing sensitivity, envelope-based measures, and scores from nonauditory cognitive tests did not significantly contribute to the variability in spatial release from speech-on-speech masking for small target/masker spatial separation, while age was a significant contributor.

1. Introduction

Speech perception difficulties in noisy acoustic environments are very common among aging individuals (CHABA 1988). These difficulties occur even in the absence of hearing loss as measured by the audiogram (Füllgrabe et al. 2015; Kim et al. 2006; Ruggles et al. 2012), indicating that factors unrelated to hair-cell damage in the cochlea are important contributors. Age-related deficits in speech understanding have often been attributed to changes in neural processing at central sites of the auditory pathways and to general decline in cognitive function (Akeroyd 2008; Dubno et al. 2002; Frisina and Frisina 1997; Henry et al. 2017; Pichora- Fuller et al. 1995; Rajan and Cainer 2008; Saunders and Haggard 1992).

The discovery of diffuse loss of auditory-nerve synapses due to noise exposure (Kujawa and Liberman 2009) and healthy aging (Sergeyenko et al. 2013) in mice, subsequently confirmed in other species (e.g., Furman et al. 2013; Hickman et al. 2018; Lin et al. 2011; Möhrle et al. 2016; Valero et al. 2017; Viana et al. 2015), gave rise to the hypothesis that some decline in recognition of speech in noisy backgrounds could originate from deficits in neural coding of suprathreshold temporal information in the auditory nerve (AN). Histopathologic analyses of synaptic survival in animal models suggest that medium- and low-spontaneous-rate AN fibers are particularly vulnerable to synaptopathy (Bourien et al. 2014; Furman et al. 2013; Liberman and Liberman 2015). These types of fibers are thought to be crucial for precise coding of the temporal characteristics of medium- and high-level sounds in the presence of simultaneous acoustic backgrounds (Liberman and Kujawa 2017), as they have relatively large dynamic ranges, their responses are robust to the effects of masking (Costalupes 1985; Young and Barta 1986), and they have the superior ability to synchronize to temporal envelopes over high-spontaneous-rate fibers (Joris and Yin 1992).

The only direct evidence of cochlear synaptopathy in humans is from temporal bone studies (Makary et al. 2011; Viana et al. 2015; Wu et al. 2019). Histopathologic analyses of temporal bone tissue have shown that the loss of AN axons becomes increasingly severe with advancing age (Wu et al. 2019), although some of the observed loss might have been due to the combined effects of age and noise exposure. Given this evidence, indirect measures that are sensitive to cochlear synaptopathy in living humans should exhibit age-related deficits consistent with those expected from synaptopathy.

This study focused on the effects of age on temporal envelope processing for stimuli presented at medium-high levels. The stimuli were either broadband or presented with noise maskers to promote reliance of the collected measures on the integrity of medium- and low-spontaneous-rate AN fibers. For such stimuli, increased cochlear synaptopathy with advancing age should result in age-related degradation of the neural representation of temporal envelopes and a corresponding decline of behavioral performance. These predictions are based on experimental data from young normal-hearing listeners with varying histories of noise exposure (Bharadwaj et al. 2015; Bharadwaj et al. 2014), on results of model simulations (Keshishzadeh et al. 2020; Vasilkov et al. 2021; Verhulst et al. 2018a, b), and on physiological data from aging animals (e.g., Parthasarathy et al. 2019). In some conditions, stimuli in quiet were used to obtain reference measures that were expected to be relatively unaffected by synaptopathy.

Auditory information contained in temporal envelopes has been shown to play an important role for understanding speech (Goossens et al. 2018; Rosen 1992; Van Tasell and Yanz 1987), with slower fluctuation rates shown to contribute to speech intelligibility in quiet and in steady noise (Drullman et al. 1994; Shannon et al. 1995; Xu et al. 2005; Xu et al. 2002), and higher rates playing a greater role in the presence of competing speech (Qin and Oxenham 2003; 2005; Stone et al. 2010). Physiological measures have shown that the fidelity of speech-envelope representation in neural responses is strongly related to speech perception (Ahissar et al. 2001; Goossens et al. 2018; Peelle and Davis 2012). The important role of temporal-envelope representation for understanding speech in noise is also supported by studies showing that speech intelligibility can be successfully predicted based solely on the target-to-masker envelope power ratio (Jørgensen and Dau 2011; Jørgensen et al. 2015; Jørgensen et al. 2013). Another form of temporal information in speech and other acoustic stimuli, the temporal fine structure, which requires phase-locking of neural spikes to the fine-grained temporal representation at the output of the cochlear filters (Joris and Yin 1992; Verschooten et al. 2019), was not considered in this study.

The effects of age on measures of temporal-envelope processing and their relation to age-related changes in speech-in-noise recognition have been previously investigated (e.g., Füllgrabe et al. 2015; Prendergast et al. 2019; Schoof and Rosen 2014), but the stimuli were not always optimal for observing effects of cochlear synaptopathy. For example, amplitude modulation (AM) of pure tones or noise bands (Schoof and Rosen 2014) can be encoded by high-spontaneous-rate fibers innervating cochlear filters tuned to frequencies on the skirts of the excitation pattern. There is also evidence from animals and humans suggesting that age effects related to cochlear synaptopathy may be particularly pronounced for high modulation rates (Mepani et al. 2021; Parthasarathy et al. 2019; Purcell et al. 2004; Shaheen et al. 2015). A few recent studies in humans, which included middle-aged and older participants, reported mixed results with some finding evidence in support of cochlear synaptopathy (Grant et al. 2020; Keshishzadeh et al. 2020; Mepani et al. 2020; Mepani et al. 2021; Vasilkov et al. 2021) and others (Moore and Vinay 2019; Prendergast et al. 2019; Valderrama et al. 2018) finding no reliable support. Because of the direct evidence of age-related cochlear synaptopathy from human temporal-bones (Makary et al. 2011; Viana et al. 2015; Wu et al. 2019), it is certain that the disorder is present in humans and that it is aggravated by increasing age. A challenge with finding reliable biomarkers of cochlear synaptopathy in older adults is that even subclinical age-related cochlear hair-cell damage may confound behavioral and electrophysiological measures of synaptopathy (e.g., Keshishzadeh et al. 2020; Vasilkov et al. 2021).

In this study, a battery of psychophysical, electrophysiological, and speech-perception measures was used to assess age-related deficits in (putatively neural) temporal-envelope representations. A multivariate model was then used to investigate if changes in these measures can predict age-related deficits in spatial release from masking (SRM) for target speech presented in two-talker babble. The SRM measure was used because the ability to use spatial cues to enhance speech understanding, particularly for small azimuthal separations, requires robust temporal processing of both the temporal fine structure and the temporal envelope (Kidd et al. 2010). Our targeted population was individuals with ages spanning a wide range (from 20 through the 60s) and with normal audiograms. However, despite screening a large number of volunteers for this study, we found a relatively small number of older listeners with hearing thresholds matching those of our youngest participants. Because of that, individuals with mild high-frequency hearing loss (≤ 40 dB HL at 4 and/or 8 kHz) were included in the study and the effects of hearing threshold were statistically controlled for in the data analyses. The limitations of this approach are discussed in Section 7.3. Based on the study of Wu et al. (2019) and other studies that used animal and computational models (Furman et al. 2013; Keshishzadeh et al. 2020; Parthasarathy et al. 2019), we expected to observe deficits in temporal processing of suprathreshold stimuli with increasing age. We expected the deficits to occur for stimuli presented in noise, but not (or to a lesser extent) in quiet, and to be more pronounced for higher envelope-fluctuation rates than for lower rates. A significant relationship between such deficits in temporal-envelope processing and speech-perception measures (speech recognition, spatial release from masking), in the absence of significant contributions of cognitive impairment, would be consistent with the hypothesis that cochlear synaptopathy contributes to difficulties with speech-in-noise perception and auditory scene analysis in the aging population.

2. Psychophysical measures of temporal-envelope processing

2.1. General methods

For all psychophysical tasks, the participants were seated in a single-walled sound-attenuating booth. The stimuli were generated in Matlab (Mathworks, Natick, MA) on a PC and were played using an E22 soundcard (LynxStudio, Costa Mesa, CA) with 24-bit resolution and a sampling rate of 48 kHz. The different experimental tasks, described below, were implemented using the AFC software package (Ewert 2013) for Matlab. For the monaural tasks (all except interaural envelope-time discrimination and speech recognition), the stimuli were delivered to the left ear, and for the binaural tasks they were delivered to both ears, through Sennheiser HD650 headphones (Sennheiser, Old Lyme, CT).

2.2. Participants

A total of 61 adults (22 males, 38 females, 1 other) with a mean age of 45.6 years (range 20 – 66 yrs) were recruited from the University of Minnesota and surrounding community. All data analyses in this study were performed with the participants divided into three age groups, youngest (20–35 yrs), middle (36–50 yrs), and oldest (51–66 yrs), and all the collected measures were compared across these age groups. Note that the labels of our age brackets describe the three age groups in relation to each other and are not based on gerontological conventions. The groups were selected to split the participants into three approximately equal-size samples along the age continuum for ease of data analyses. The division was also in part motivated by the fact that individuals in their 40s and 50s often report difficulties understanding speech in noisy backgrounds even when the audiogram is normal. For each participant, audiometric thresholds were measured for both ears at octave frequencies between 0.25 and 8 kHz, using a calibrated audiometer (Madsen Conera, GN Otometrics). Normal hearing was defined by pure-tone air-conduction thresholds ≤ 20 dB HL. All participants had normal thresholds for frequencies ≤ 2 kHz in both ears, but six participants in the middle group and thirteen in the oldest group had a mild hearing loss (≤ 40 dB HL) at one or two of the highest frequencies (4 and 8 kHz). All participants had symmetric hearing (across-ear differences ≤ 10 dB) at all audiometric test frequencies. Hearing thresholds at the same octave frequencies were also measured using a two-alternative forced-choice (2AFC) procedure with a two-down, one-up adaptive-tracking rule that estimates the 70.7% correct point on the psychometric function (Levitt 1971). The 2AFC procedure was used to obtain a fine-grained, unbiased measure of hearing thresholds for better statistical control of effects of hearing sensitivity. The steps in the adaptive tracking were 8 dB for the first two reversals, 4 dB for the subsequent two reversals, and 2 dB for the final eight reversals. The tone levels at the final eight reversals were averaged to calculate a threshold for each run. Two estimates of hearing threshold were averaged to obtain the final threshold estimate in dB SPL at each audiometric frequency. Thresholds averaged across participants within each age group are shown by different symbols (and lines connecting them) in Fig. 1.

Fig. 1.

Hearing thresholds in dB SPL measured using a 2AFC adaptive-tracking procedure averaged within each of the three age groups specified in the legend. The error bars indicate ± one standard error (SE) of the mean.

Despite all of the participants having clinically normal audiometric thresholds for frequencies ≤ 2 kHz, thresholds measured with the 2AFC procedure averaged within each age group show progressive loss of hearing sensitivity with increasing age for all frequencies between 0.25 and 8 kHz, more so for frequencies above 2 kHz. A mixed-design analysis of variance (ANOVA), with tone frequency as a within-subjects factor and age group as a between-subjects factor, showed a significant effect of frequency [F(5, 290) = 43.2, p < 0.001], a significant effect of age group [F(2, 58) = 17.7, p < 0.001], and a significant interaction between frequency and age group [F(10, 290) = 5.7, p <0.001]. Post hoc pairwise comparisons showed significant differences in hearing sensitivity between the youngest and middle age groups (p = 0.012), youngest and oldest groups (p < 0.001), and middle and oldest groups (p = 0.011).

Tympanometry was performed for each participant to ensure normal middle-ear function (type-A tympanogram) using a calibrated tympanometer (MT10, Interacoustics, DK). Not all individuals who consented to participate in this study completed the entire protocol. Table 1 shows the total number of participants, the sample sizes, the mean ages for each group, the means of hearing thresholds averaged across the entire audiometric range, 0.25 – 8 kHz (mean ALLAVG), and the means of hearing thresholds averaged across the three highest frequencies, 2, 4, and 8 kHz (mean HFAVG) for each age group for each experiment. A total of 27 recruited participants were able to take part in all experiments. Participants received monetary compensation after each experimental session. Each participant provided informed written consent before testing began and the experimental protocols were approved by the Institutional Review Board at the University of Minnesota.

Table 1.

The sample size (female, F), the mean age, the mean of hearing thresholds averaged across frequencies 0.25 – 8 kHz (ALLAVG), and the mean of hearing thresholds across frequencies 2 – 8 kHz (HFAVG), for the three age groups, youngest (Y), middle (M), and oldest (O), for each experimental task listed in column 1. The age range of participants (20 – 66 years) was same for each task.

Experimental task	Sample size	Mean age (years)			Mean ALLAVG (dB SPL)			Mean HFAVG (dB SPL)
		Y	M	O	Y	M	O	Y	M	O
AM Detection	61 (38 F)	26.8	46.7	63.4	4.8	13.1	21.1	6.0	18.3	30.1
Gap Detection	53 (32 F)	25	43	60.5	5.1	14.2	23.4	10.9	20.1	31.3
envITD	27 (20 F)	25.8	44.1	60.8	5.2	16.3	25.1	10.3	19.2	30
Speech	58 (36 F)	26.6	46.5	63.2	4.9	14	22.2	9	19.2	31
EFR	45 (33 F)	25	44.6	62.8	4.9	14	22.2	6.5	17.4	30.2
All tasks	27 (20 F)	25.8	44.1	60.8	5.2	16.3	25.1	10.3	19.2	30

2.3. Stimuli and procedures

2.3.1. Detection of AM of a noise carrier

Thresholds for detecting sinusoidal AM imposed on a Gaussian-noise carrier were measured for three modulation rates, 100, 300, and 600 Hz. The noise carrier, with a bandwidth from 0.1 to 10 kHz, was used to limit the role of high-spontaneous-rate fibers in coding intensity fluctuations on the skirts of the excitation pattern. Detecting AM for noise carriers is limited by modulation masking from the inherent envelope fluctuations in the carrier (Dau et al. 1997a; b). Because of that performance may be more susceptible to the effects of cochlear synaptopathy than detecting AM for tonal carriers with flat envelopes, as undersampling of a target AM should be more detrimental in the presence of other envelope fluctuations (Lopez-Poveda 2014). However, the use of a broadband noise carrier could introduce confounding effects for participants with high-frequency hearing loss, an issue that is discussed below (see Section 2.4.1). Modulation rates ≥ 100 Hz were chosen because some previous studies showed age effects on AM responses for high, but not for low rates (Grose et al. 2009; Purcell et al. 2004).

A 2AFC procedure, combined with a three-down, one-up tracking rule, was used to estimate the threshold corresponding to the 79.4 % correct point on the psychometric function (Levitt 1971). In one interval, selected at random, the noise was modulated throughout its duration while the other interval contained an unmodulated noise. The stimuli in the two intervals were scaled to have equal root-mean-square amplitudes to reduce loudness cues (Viemeister 1979) and were presented at 70 dB SPL. The noise carrier had a duration of 250 ms including 10-ms onset/offset ramps, and the two observation intervals were separated by 300 ms of silence. The observation intervals were marked by flashing colored boxes on a computer screen. In each trial, participants were asked to choose the interval containing the AM and provide their response by a keypress or via a mouse click. Visual feedback indicating the correct response was provided after each trial. The tracking variable was the modulation depth expressed in dB [20 log (m)], where m is the modulation index with a value between 0 and 1. The modulation depth was changed in 4-dB steps for the first two reversals, 2-dB steps for the next two reversals, and 1-dB steps for the final eight reversals. The AM-detection threshold for a single run was calculated by averaging modulation depths (in dB) at the last eight reversals. Thresholds from three runs were averaged to compute the final threshold estimate. When the adaptive-tracking procedure called for a modulation depth greater than 0 dB (100% AM), the track was aborted and another measure was started. This happened for 10 listeners for only one of the three modulation rates (mostly for 600 Hz). All these listeners were able to successfully complete three runs after one aborted track.

2.3.2. Gap detection for tones in noise

Gap detection thresholds were measured at two frequencies, 2 and 4 kHz, for tones embedded in one-octave wide Gaussian noise. For each tone, the noise was geometrically centered at the tone’s frequency. The two frequencies were selected because no participants had thresholds > 20 dB HL at 2 kHz, and some participants in the middle and oldest groups had audiometric thresholds between 20 and 40 dB HL at 4 kHz. The ability to detect a short gap in a tone relies on both temporal and intensity resolution (Moore et al. 1989). In the presence of outer-hair-cell damage, effects of cochlear synaptopathy on gap-detection threshold (putatively, impaired temporal resolution) could be offset by effects of linearized cochlear response (enhancement of a dip in stimulus intensity). This potential trade-off could result in no age effect on gap detection. Since hair-cell damage in middle and older participants occurred (or, at least, was more severe) at 4 kHz, the trade-off would more likely occur at 4 than at 2 kHz. The levels of the tones and noise bands were 75 and 65 dB SPL, respectively, resulting in a signal-to-noise ratio (SNR) of 10 dB. A three-interval, three-alternative forced-choice (3AFC) procedure, combined with a three-down, one-up adaptive tracking technique, was used to measure thresholds corresponding to the 79.4 % correct point on the psychometric function (Levitt 1971). The tonal markers before and after the gap were 250-ms long including 1-ms hanning onset and offset ramps on either side of the gap. The noise was 700-ms long including 10-ms ramps. In each observation interval, the first marker started 50 ms after the noise onset. In the non-signal intervals, the markers were presented in sequence with a 0-ms gap between the end of the offset ramp of the first marker and the beginning of the onset ramp of the second marker. The off/on ramping resulted in a brief (1-ms interval between half-amplitude points) intensity dip in the non-signal intervals that was not detectable in the noise masker. The use of a 3AFC task would have prevented confusion if the dip had been detected, as participants could simply select the interval that sounded different from the other two. In the signal interval, a silent gap was inserted between the markers. The silent gap was partially filled with the noise masker. The gap duration was changed by a factor of 2 for the first two reversals, $\sqrt{2}$ for the next two reversals, and $\sqrt[4]{2}$ for the final eight reversals. A track would have been aborted if the adaptive procedure called for a gap duration that would cause the post-gap marker to extend beyond the duration of the noise masker, but none of the participants reached that limit. Threshold from a single run was calculated as the geometric mean of gap durations at the last eight reversals. The final estimate was the geometric mean of thresholds from three runs.

2.3.3. Interaural envelope time difference (envITD)

Sensitivity to envITDs was measured using a 2AFC procedure combined with a two-down, one-up adaptive track estimating threshold at the 70.7% correct point on the psychometric function (Levitt 1971). The measurements were performed using two carrier frequencies, 2 and 4 kHz and two AM rates, 40 and 100 Hz, of full (100%) sinusoidal AM. The two modulations rates were used because data from animal studies suggest that greater effects of cochlear synaptopathy should be observed for high rather than low rates of envelope fluctuation (Parthasarathy et al. 2016; Shaheen et al. 2015), since slower fluctuations may be reconstructed by compensatory neural (plastic) mechanisms at central sites (Chambers et al. 2016; Parthasarathy et al. 2019). Based on these findings, it was hypothesized that stronger age effects on envITD thresholds would be observed for the 100- than for the 40-Hz AM.

In each interval, a sequence of four 500-ms AM carriers, separated by a 20-ms silent gap, was presented to both ears. The carrier duration included 100-ms hanning onset and offset ramps. In the signal interval, selected at random, the first and third tones had AM with the same phase in the two ears (diotic AM tones), while the AM in the second and fourth tones had an AM phase difference between the two ears, resulting in an envITD (an ABAB sequence similar to that used by Hopkins and Moore 2010). Because relatively high modulation rates were used for the suprathreshold envITDs, the stimulus was perceived as changing between a narrowly centered intracranial sound image to a broader image spreading toward both sides of the head over the course of the four tone bursts in an observation interval. The non-signal interval consisted of four diotically presented AM tones (AAAA sequence), resulting in a fixed (centered) sound image. The envITD thresholds were measured in quiet and in the presence of a notched-noise masker. It was hypothesized that age effects would be stronger for carriers presented in a notched noise than for carriers in quiet (Bharadwaj et al. 2015). The noise had a bandwidth from 0.1 to 10 kHz and contained a spectral notch with a width of 0.2 × carrier frequency, geometrically centered on the carrier frequency. The notched noise was used to mask low-level portions on the skirts of the excitation pattern produced by the modulated tones. The geometric centering was used because of the asymmetric shape of excitation patterns in the cochlea, with excitation level decreasing more steeply on the low-frequency side (Glasberg and Moore 1990). The notched noise started 100 ms before the first observation interval and ended 100 ms after the end of the second observation interval. Different samples of noise were generated for each trial separately for the left and right ears to minimize the effects of noise lateralization on envITD thresholds. The two observation intervals were separated by a 500-ms gap. For thresholds measured in quiet, the gap contained silence and for thresholds in noise, the notched noise continued throughout the gap.

The carrier level was 70 dB SPL and the notched noise had an overall level of 50 dB SPL. Listeners were asked to choose the interval in which they perceived movement of the sound image in their head across the four stimulus presentations. The observation intervals were marked by flashing colored boxes and visual feedback indicating the correct response was provided after each trial. The adaptive tracking variable was the log-transformed interaural envelope phase difference. The track started with the phase difference set to π. The phase difference was changed by a factor of 1.25² for the first four reversals and 1.25 for the remaining eight reversals. Thresholds from a single run were estimated by calculating the geometric mean of the interaural envelope phase differences at the final eight reversals. Three single-run threshold estimates were averaged (geometrically) to calculate the final threshold. The interaural phase difference at threshold was then converted to the interaural time difference, envITD. All participants performed one practice run in each experimental condition before data collection began to ensure they understood the concept of intercranial image changes (broadening and narrowing) in the signal interval.

2.4. Results and discussion

2.4.1. AM detection

Sixty-one participants completed the AM-detection experiment. The mean thresholds for the three age groups are shown in Fig. 2. As expected, the three groups exhibited an increase (worsening) of AM detection threshold with increasing modulation rate between 100 and 600 Hz, consistent with the low-pass characteristic of the temporal modulation transfer function for broadband noise carriers (Viemeister 1979).

Fig. 2.

AM-detection thresholds for a broadband noise carrier averaged within each of the three groups, for three modulation rates. The error bars indicate ± 1 SE. The abbreviation, n.s. (here and in all other figures), denotes a non-significant difference between age groups.

A mixed-design ANOVA was performed on the AM-detection thresholds with modulation rate as a within-subjects factor and age group as a between-subjects factor. The average of the hearing thresholds, measured in dB SPL, for frequencies of 2, 4, and 8 kHz, HFAVG, was used as a covariate. These frequencies were included in the covariate because hearing thresholds diverged between the three age groups in the high-frequency region, and because high modulation rates are processed in auditory filters tuned to high frequencies (Dau et al. 1997a; Kohlrausch et al. 2000). Because Mauchly’s test showed that the assumption of sphericity was violated, a Greenhouse-Geisser correction was applied to the within-subjects effects. There was a significant effect of modulation rate [F(1.6, 93.4) = 132.81, p < 0.001], a significant effect of HFAVG [F(1, 57) = 5.59, p = 0.022], but no significant effect of age group after controlling for HFAVG [F(2, 57) = 1.55, p = 0.22]. No significant interactions were observed between AM rate and HFAVG [F(1.6, 93.4) = 1.70, p = 0.187] or between AM rate and age group [F(3.3, 93.4) = 2.10, p = 0.1].

The lack of an effect of age group on sensitivity to high-rate AM contrasts with significant age effects reported by Kumar and A. V. (2011). The age of the participants in that study spanned a wider range (20 – 85 yrs) than for our participants (20 – 66 yrs). For a modulation rate of 200 Hz, the highest rate used, Kumar and A. V. (2011) found that AM-detection thresholds deteriorated with increasing age between the youngest group (20–30 yrs) and the middle-aged group (41–50 yrs), with no further decline for the older group (50 – 85 yrs). The main difference between our results and those of Kumar and A. V. (2011) is that the middle and oldest groups in our study performed much better than the participants older than 41 yrs in their study. A possible reason for this difference is that all participants below the age of 65 yrs tested by Kumar and A. V. (2011) had hearing thresholds below 15 dB HL at all audiometric frequencies, whereas some of our middle and oldest participants had hearing loss at higher frequencies. Sensorineural hearing loss could potentially result in improved sensitivity to AM as outer hair cell damage is associated with broadening of the cochlear filters and linearization of the cochlear input-output function (Glasberg and Moore 1989; Nelson et al. 2001; Vasilkov et al. 2021). Although better AM-detection thresholds have been observed for hearing-impaired than for normal-hearing adults for pure-tone carriers (Ernst and Moore 2012; Füllgrabe et al. 2003; Jerger 1962), hearing impairment has been shown to either have no effect (Bacon and Gleitman 1992; Moore et al. 1992b) or to be detrimental for detecting AM of noise carriers (Bacon and Viemeister 1985; Takahashi and Bacon 1992). The detrimental effect has been explained in terms of a limited stimulus bandwidth at the output of the cochlea, and consequently a smaller number of filters over which AM information could be integrated across the filter outputs (Bacon and Viemeister 1985). Linearization of cochlear responses also could have a different effect for tonal versus noise carriers. For noise carriers, both the signal AM and the inherent envelope fluctuations in the noise carrier are affected by the more linear processing. Because AM-detection thresholds for noise carriers are limited by modulation masking by the inherent envelope fluctuations in the carrier (Dau et al. 1997a; b), the benefit from more linear cochlear processing could be nullified. In contrast, tonal carriers have a flat envelope, and thus a more linear cochlear response would facilitate AM detection. On the other hand, broadening of cochlear filters results in a smoother envelope of the noise carrier at the output of the filters. This smoothing could decrease the amount of modulation masking of the signal AM in impaired ears compared to healthy ears, thereby resulting in the better performance for the middle and oldest groups in this study.

To examine this possibility, we compared AM-detection thresholds across the three age groups for a subset of participants whose thresholds in dB SPL were matched within a 2-dB range at all frequencies. The subset consisted of 18 participants (6 per age group). For this subset, the pattern of results (data not shown) was very similar to that for the original sample with nearly identical mean modulation depths at threshold for each age group. A mixed-design ANOVA showed no significant effect of age group [F(2, 14) = 1.22, p = 0.33] and no significant interaction between AM rate and age group [F(4, 28) = 1.46, p = 0.24] for the three groups matched for hearing sensitivity. Not surprisingly, there was also no effect of HFAVG [F(1, 14) = 0.004, p =0.95], confirming that matching hearing thresholds removed any effects that could be attributed to differences in hearing sensitivity. Notably for this subset of participants, AM-detection thresholds for our middle and oldest groups were still substantially lower than those for the participants aged > 41 yrs in Kumar and A. V. (2011) for the 200-Hz rate (most comparable to the modulation rates used in the current study), indicating that hearing loss of our participants may not entirely account for the different findings.

Takahashi and Bacon (1992) also observed reduced sensitivity to AM for a noise carrier for three groups of older participants (in their 50s, 60s, and 70s) compared with a young group (21–33 yrs). The difference between the groups increased with increasing modulation rate. All three groups of older participants had on average some hearing loss at 4 and 8 kHz, possibly limiting the available bandwidth of the modulated noise carrier at the output of the cochlea. In contrast, the middle and oldest groups of participants in the current study did not differ in their sensitivity to AM across the three modulation rates used (Fig. 2) despite the fact that the spectrum level of the noise carrier was slightly lower than in Takahashi and Bacon (~30 dB SPL vs. ~35 dB SPL), and the two oldest groups in their study had on average the same amount of hearing loss as the oldest group in our study.

To further explore effects of age and hearing sensitivity, correlations between AM-detection thresholds and age, and between AM-detection thresholds and HFAVG, were calculated for data pooled across the three age groups separately for each modulation rate. Partial correlations controlling for age and HFAVG were also calculated. The correlation coefficients are shown in Table 2. The correlations with age and HFAVG were statistically significant only for the 100-Hz AM rate at the Bonferroni-corrected significance level (α=0.008, for six comparisons). The partial correlation with HFAVG remained significant but only at the uncorrected level. The lack of consistent significant age effects on AM detection may be due to a tradeoff between enhanced AM representation due to outer hair-cell loss and limited overall bandwidth of the stimulus at the output of the cochlea of participants with high-frequency hearing loss, including frequencies above the audiometric range. However, it cannot be ruled out that the data show a tradeoff between enhancement of AM due to hearing loss and degraded AM representation due to cochlear synaptopathy. For the purpose of the current study, the AM-detection task was performed to determine if AM-detection thresholds for high modulation rates applied to a noise carrier can significantly predict the ability to use small spatial cues to improve speech-in-speech recognition. This is discussed below.

Table 2.

Coefficients of Pearson’s product correlations between AM-detection thresholds and age, and AM-detection thresholds and HFAVG (2 – 8 kHz), performed separately for each modulation rate. Two rightmost columns show partial correlations with the controlled variable in parentheses. Correlations that are significant after correcting for multiple comparisons are shown in bold font (p < 0.0167).

AM rate [Hz]	Age	HFAVG	Age (HFAVG)	HFAVG (Age)
100	−0.34	−0.40	−0.15	−0.26
300	−0.03	−0.27	0.16	−0.31
600	−0.04	−0.19	0.08	−0.20

2.4.2. Gap detection for tones in noise

Fifty-three participants completed the gap-detection task for tones presented in noise. The average gap-detection thresholds for the three age groups are plotted in Fig. 3. The left and right sets of bars show the gap thresholds for the 2- and 4-kHz tones, respectively. As is evident in the figure, listeners in all three age groups were able to detect shorter gaps at 4 kHz than at 2 kHz.

Fig. 3.

Gap-detection thresholds for 2- and 4-kHz tones presented in one-octave noise bands, averaged within each of the three age groups. The error bars indicate ± 1 SE. Significant differences between age groups are denoted by asterisks (*** p < 0.001, and * p < 0.05).

A mixed-design ANOVA performed on log-transformed gap-detection thresholds with frequency of the tonal marker as a within-subjects factor, age group as a between-subjects factor, and HFAVG (defined above) as a covariate, showed a significant effect of marker frequency [F(1, 49) = 21.32, p < 0.001]. There was also a significant effect of age group [F(2, 49) = 5.64, p = 0.006], but no significant effect of HFAVG [F(1, 49) = 0.19, p = 0.67]. There was no significant interaction between marker frequency and age group [F(2, 49) = 0.21, p = 0.82] or between marker frequency and HFAVG [F(1, 49) = 1.35, p = 0.25]. The Games-Howell post hoc test was used for pairwise comparisons, separately for each marker frequency. For the 2-kHz marker, none of the pairwise comparisons showed significant age group differences (all p > 0.05). For the 4-kHz marker, there was a significant difference in gap-detection thresholds between the youngest and middle (p = 0.038) and the youngest and oldest groups (p < 0.001), but not between the middle and oldest groups (p > 0.05).

Treating age as a continuous variable, Pearson correlations between gap-detection thresholds and age, and gap thresholds and HFAVG, were calculated separately for each tone frequency. As shown in Table 3, neither age nor HFAVG were significantly correlated with gap-detection thresholds for the 2-kHz tone. For the 4-kHz marker, the thresholds were significantly correlated with age and the correlation remained significant after controlling for HFAVG at the Bonferroni-corrected significance level (α = 0.0125, for four comparisons). In sum, the results of the correlational analyses show that age but not high-frequency hearing sensitivity contributed to a decline in the detectability of temporal gaps in a 4-kHz tone presented in noise.

Table 3.

Coefficients of Pearson’s product correlations between gap-detection thresholds and age, and HFAVG (2 – 8 kHz). Correlations were performed separately for the 2- and 4-kHz tonal markers of the gap. The two rightmost columns show partial correlations with the controlled variable listed in parentheses. Bonferroni-corrected significant correlations are shown in bold font (p < 0.025).

Frequency [kHz]	Age	HFAVG	Age (HFAVG)	HFAVG (Age)
2	0.27	0.06	0.27	−0.08
4	0.56	0.30	0.50	0.04

Overall, the data are consistent with a large number of studies which reported increased gap-detection thresholds with advancing age for a variety of stimuli including pure tones (Moore et al. 1992a; Schneider and Hamstra 1999), noise bands (Humes et al. 2009; Humes et al. 2010; Snell 1997; Snell and Frisina 2000; Snell et al. 2002), broadband noise (Kumar and A. V. 2011; but cf. Schoof and Rosen 2014), and speech (Pichora-Fuller et al. 2006). Deterioration of gap-detection performance in middle aged (> 40 yr) listeners compared to young adult listeners has been shown for low-frequency tones (Grose et al. 2006) and for broadband-noise markers (Kumar and A. V. 2011). Our data for the 4-kHz marker show the same pattern, i.e., gap-detection thresholds were significantly worse for the middle and oldest groups than for the youngest group. For the 2-kHz marker, there was a trend in the same direction (see Fig. 3) that did not reach significance. The statistical analyses show that the decline in the ability to detect brief gaps in stimuli presented at suprathreshold levels in noise with advancing age is independent of the hearing loss that is often associated with aging (e.g., Humes et al. 2010).

2.4.3. Interaural time difference in stimulus envelope

Thresholds for detecting envITDs for two carrier frequencies (2 and 4 kHz) and two modulation rates (40 and 100 Hz) were measured in quiet and in noise for 27 listeners. The log-transformed envITDs at threshold averaged within each age group are shown in Fig. 4. The left and right panels show data for the 40- and 100-Hz AM rates, respectively. In each panel, the left and right set of bars show the thresholds for the 2- and 4-kHz carriers, respectively, in quiet and in noise.

Fig. 4.

Thresholds for detecting interaural time differences in envelope (envITD), for the three age groups. The error bars indicate ± 1 SE. Data for the 40- and 100-Hz AM are shown in the left and right panels, respectively. Thresholds were measured for 2- and 4- kHz carriers in quiet (Q) and in one-octave noise (N). The horizontal dotted lines indicate the envITD corresponding to an interaural envelope phase difference of 180 deg, for the 40-Hz (left panel) and 100-Hz (right panel) AM rates.

A mixed-design ANOVA was performed on the log-transformed envITDs for each modulation rate. The two modulation rates were considered separately since comparable phase shifts for the 40- and 100-Hz rates correspond to different envITDs. In both cases, carrier frequency and condition (quiet/in noise) were within-subjects factors, age group was a between-subjects factor, and HFAVG was a covariate. For the 40-Hz AM rate, there was no significant effect of carrier frequency [F(1, 23) = 0.726, p =0.403] and no significant effect of condition [F(1, 23) = 1.31, p = 0.299]. There was no significant effect of age group [F(2, 23 = 1.88, p = 0.175] or HFAVG [F(1, 23) = 0.451, p = 0.508]. None of the two- or three-way interactions were significant (all p > 0.05). For the 100-Hz AM rate, the effect of carrier frequency was significant [F(1, 23) = 15.09, p = 0.001], indicating performance being worse for the 4-kHz than for the 2-kHz carrier. The effect of condition (quiet/in noise) was also significant [F(1, 23) = 5.61, p = 0.027], indicating that the presence of noise had an adverse effect on envITD thresholds. However, there was no significant effect of age group [F(2, 23) = 0.87, p = 0.432], no significant effect of HFAVG [F(1, 23) = 0.014, p = 0.907], and no significant two- or three-way interactions of the two factors with carrier frequency and condition (all p > 0.05).

For all participants pooled together, there were no significant correlations between age and envITD thresholds, between HFAVG and envITD thresholds, and no significant partial correlations with either of the two variables with the other variable controlled for (all p > 0.05). This was true for all conditions tested (two carrier frequencies, two modulation rates, in quiet and in noise). Age-related cochlear synaptopathy was expected to produce significant effects of age for stimuli in noise, particularly for the higher modulation rate (Bharadwaj et al. 2015; Shaheen et al. 2015). The pattern of results was not consistent with these expectations.

Consistent with our findings, Prendergast et al. (2019) showed no age effects on sensitivity to envITDs for a carrier frequency of 4 kHz, equal to one of the carrier frequencies used in the current study. Notably, Prendergast et al. (2019) used an even higher modulation rate (255 Hz), for which previous studies have found deficits in monaural AM processing (Kumar and A. V. 2011; Purcell et al. 2004). However, discrepant findings have been reported regarding age effects on sensitivity to envITDs at low carrier frequencies (250 and 500 Hz) and relatively low modulation rates (20 – 50 Hz). For low-frequency carriers, King et al. (2014) found a significant positive correlation between envITD and age for participants aged 18 – 83 yrs, whereas Moore et al. (2018) found no significant age effects when comparing two groups on the extreme ends, 19 −27 vs. 62 – 78 yrs. The lack of significant correlation with hearing thersholds in the frequency range corresponding to the carrier frequencies in the current study and in the study of King et al. (2014) contrasts with the significant contribution of hearing sensitivity to envITDs reported by Prendergast et al. (2019). One reason for the discrepant findings may be the larger sample size (N =156) in Prendergast et al. (2019) than in our (N = 27) and the King et al. (2014) (N=46) studies. The correlations reported by Prendergast et al. (2019) showed that hearing sensitivity predicted only about 11% of the variance in envIPD thresholds, a small effect that perhaps could not be detected with the smaller samples given the inherent variability in envITD thresholds both within and across participants.

3. Electrophysiological envelope following responses (EFRs)

3.1. Participants

Electrophysiological EFRs were measured for 45 (33 female) of the 61 initially recruited participants. The remaining participants were unavailable for testing due to time constraints. Most of the participants completed at least two of the psychophysical tasks described in Section 2.

3.2. Stimuli and procedure

EFRs were measured in response to a series of 437.52-ms long tones modulated in amplitude at a rate of 91.42 Hz. Of the 45 participants, 21 (6 youngest, 8 middle, and 7 oldest) with normal hearing at all audiometric frequencies were tested using a carrier frequency of 4 kHz. Because eight participants in the oldest group had a mild hearing loss at 4 kHz, the remaining 24 participants (8 youngest, 8 middle, and 8 oldest) were tested using a carrier frequency of 2 kHz. This approach was different from the one taken in the psychophysical tasks in which all participants were tested using both 2 and 4 kHz because we wanted to limit the EEG measurements to one (~2-hour) session to reduce participant attrition. The AM tones were presented in quiet and in notched noise. For the AM tones in quiet, the EFRs were measured for modulation depths of −8 and 0 dB [20log(m)], while for the AM tones in noise, the EFRs were measured for modulation depths of −8, −4, and 0 dB. It was hypothesized that for AM tones in quiet, age-related cochlear synaptopathy would have little or no effect on EFRs for both modulation depths, as high-spontaneous-rate AN fibers could reliably encode the presence of AM. In noise, the age-related synaptopathy should have a greater effect for smaller modulation depths. This is because for smaller modulation depths, sound intensity during the valleys of the AM tones is relatively high, and thus the modulation should be coded predominantly by medium- and low-spontaneous-rate fibers (Bharadwaj et al. 2014; but see Carney 2018). The masking noise spectrum extended from 0.1 to 10 kHz with a spectral notch geometrically centered on the carrier frequency. The notch width was 0.2 times the carrier frequency. In each condition, the AM tones were presented 1000 times with an average inter-stimulus interval of 550 ms with a small time jitter distributed uniformly over a 20-ms interval around the mean. The stimuli were presented with positive polarity in half of the trials and negative polarity in the other half. The order of the five test conditions (quiet x 2 modulation depths, in noise x 3 modulation depths) was randomized across participants. The AM tones were presented at 70 dB SPL, and the notched noise had an overall level of 60 dB SPL.

Participants were seated in an electrically shielded double-walled sound-attenuating booth and neural responses were recorded using a 64 channel EEG system (Biosemi Active II system). Participants wore a cap (Easy Cap, Falk Minow Services) containing 64 silver-chloride scalp electrodes. A reference electrode was placed on the mastoid on the side of the test ear, and two additional ocular electrodes were used to record eye movement and eyeblink activity. The voltage DC offset was maintained below ± 20 mV for all electrodes. The neural activity was recorded at a sampling rate of 4096 Hz. A desktop computer was used to present and trigger the stimuli, using the Biosemi software, and store the data. The stimuli were generated with a sampling rate of 44.1 kHz and were routed to a Tucker-Davis Technologies System 3 (Gainesville, FL) for monaural presentation via an ER-1 insert earphone with a foam ear tip (Etymotic Research, Elk Grove Village, IL). The ear used for presentation was the same as that used for the monaural psychophysical tasks described in Section 2.1. To avoid boredom during the measurements, participants watched a silent captioned movie of their choice.

The pre-processing and averaging of the EEG recordings was performed using the EEGLAB toolbox (Delorme and Makeig 2004). The raw waveforms were first downsampled to 1024 Hz, re-referenced to the test-ear mastoid, and band-pass filtered over the range 1 to 350 Hz using a 4th-order Butterworth filter. A zero phase shift was achieved using Matlab’s filtfilt function. The continuous EEG time waveforms were divided into epochs, from −100 ms before stimulus onset to 500 ms post onset. The 100-ms pre-stimulus interval was used for baseline correction. Independent Component Analysis (ICA) was conducted to isolate and remove artifacts related to eye movements and eyeblinks (Jung et al. 2000). Discrete Fourier transforms of the pre-processed EEG waveforms from individual trials were performed to obtain the phase spectrum. For each electrode, the phase locking value (PLV) to the AM in the stimulus was calculated by averaging the phases at each frequency across 800 randomly selected samples from individual trials (Zhu et al. 2013). A bootstrapping technique, as described by Zhu et al. (2013), but for 800 trials, was used to estimate the average PLV and the noise floor. Based on this analysis, a PLV was considered significant when it exceeded 0.045. The subset of 35 electrodes that yielded the largest PLVs in the test ear hemisphere was selected to calculate the average PLV that was used as the final estimate for each subject and each condition. These PLV estimates were used for statistical comparisons to test for the effects of age-related cochlear synaptopathy on neural synchrony to AM.

3.3. Results and discussion

All the estimated PLVs were above the limit corresponding to the 95th percentile of the noise floor distribution and thus, all were included in statistical analyses. The PLVs averaged separately across participants who completed the experiment for the 2- and 4-kHz carriers are shown in the left and right panels of Fig. 5, respectively. The different color bars show the PLVs for the different age groups. The presence of the notched noise substantially reduced PLVs compared with those observed for the AM tones in quiet.

Fig. 5.

EFR phase-locking values for 91.4-Hz AM of a 2-kHz carrier (left panel) and a 4-kHz carrier (right panel). The error bars indicate ± 1 SE of the mean. The EFRs were measured for two modulation depths, −8 and 0 dB, in quiet (Q), and for three modulation depths, −8, −4, and 0 dB, in noise (N). The asterisks indicate significant age-group differences (* p < 0.05).

Mixed-design ANOVAs were performed separately on the PLVs measured using the 2- and 4-kHz carriers. The five test conditions (quiet/noise and 2/3 modulation depths) were used as within-subjects factors, age group was used as a between-subjects factor, and HFAVG was used as a covariate. Because Mauchly’s test showed that the assumption of sphericity was violated for both carrier frequencies, the Greenhouse-Geisser correction was applied in both cases. For the 2-kHz carrier, the ANOVA showed significant effects of condition [F(2.23, 80) = 16.92, p < 0.001] and age group [F(2, 20) = 3.91, p = 0.037], but no significant effect of HFAVG [F(1, 20) = 1.32, p = 0.265]. There was no significant interaction between condition and age group [F(4.47, 80) = 1.90, p = 0.121] or between condition and HFAVG [F(2.23, 80) = 0.64, p = 0.548]. The Games-Howell test showed a significant difference between the youngest and oldest groups for the 0-dB modulation depth in quiet (p = 0.048). In all other conditions, none of the pairwise group comparisons was significant. For the 4-kHz carrier, the ANOVA showed significant effects of condition [F(2.36, 68) = 8.66, p < 0.001] and age group [F(2,17) = 6.28, p = 0.009], but no significant effect of HFAVG [F(1, 17) = 0.008, p = 0.931]. There was no significant interaction between condition and age group [F(4.71, 68) = 1.76, p = 0.146] or between condition and HFAVG [F(2.36, 68) = 0.77, p = 0.488]. The Games-Howell test showed significant differences between the youngest and middle groups (p = 0.01) and between the youngest and oldest groups (p = 0.031) for the 0-dB AM depth in quiet, and between the youngest and middle groups (p = 0.017) and the youngest and oldest groups (p = 0.009) for the −8-dB AM in noise.

Correlations between EFR PLVs pooled across participants from all age groups and the two carrier frequencies, and age, and HFAVG were calculated separately for each of the five conditions. None of the correlations and none of the partial correlations with either variable was significant after controlling for the other variable (all p > 0.05).

One reason for the lack of age effects on EFRs could be that the modulation rate used in this study was not high enough to reveal deficits in older individuals (Purcell et al. 2004). Other studies using similar rates (~100 Hz) and lower modulation depths also reported no effects of age (Bharadwaj et al. 2015; Prendergast et al. 2019), although Bharadwaj et al. (2015) only tested listeners younger than 40 yrs. However, noise exposure, which may cause cochlear synaptopathy regardless of age, was found to be positively correlated with the slope of a function relating EFR spectral magnitudes to modulation depth in the studies of Zhu et al. (2013) and Bharadwaj et al. (2015). They argued that the slope-based measure was a better summary measure than the EFR magnitude itself as it provided normalization that removed influences of non-auditory factors on EEG recordings (Mitchell et al. 1989), thus facilitating between-group comparisons of AM processing. In both these studies, the slopes were steeper for individuals for whom behavioral measures of temporal envelope processing showed patterns consistent with the presence of cochlear synaptopathy. In apparent contrast, Garrett and Verhulst (2019) observed shallower EFR slopes for an older group than for a young group of normal-hearing participants. Garrett and Verhulst (2019) suggested that high-frequency hearing loss in older participants may have offset the effect expected from cochlear synaptopathy. However, a large-scale study of Prendergast et al. (2017a) also found no relationship between EFR slopes and noise exposure. In the present study, the data in quiet were collected for only two modulation depths, yielding highly variable slope estimates. However, we fitted straight lines to PLVs for the three modulation depths for the AM tones presented in the notched noise. ANOVAs conducted separately on the slopes for the 2- and 4-kHz carriers showed no effect of age group [F(2, 23) = 0.44, p = 0.653, for the 2-Hz carrier, and F(2, 20) = 0.74, p = 0.491, for the 4-kHz carrier] or HFAVG [F(2, 23) = 3.11, p =0.093 for the 2-kHz carrier, and F(2, 20) = 0.52, p = 0.481 for the 4-kHz carrier]. Overall, except for isolated conditions we found no effects of age or HFAVG on either the PLVs or the slopes of the functions relating PLVs to modulation depth.

4. Speech perception and spatial release from masking

4.1. Participants

Fifty-eight listeners (36 females) completed the speech experiment. Twenty-seven of them performed all the experiments in this study. The rest performed at least one other experiment.

4.2. Stimuli and procedure

The speech stimuli were sentences from the Institute of Electrical and Electronics Engineers (IEEE 1969) corpus, spoken by a female with a mean voice fundamental frequency (F0) of ~ 180 Hz. The background speech consisted of a mixture of two female voices that had either the same mean voice F0 as the target (same-F0 condition) or were processed using the PRAAT software package (Boersma and Weenink 2010) to yield one background speaker with an F0 that was three semitones below, and one with an F0 that was three semitones above that of the target (different-F0 condition). The spatial locations of the target and background speakers were simulated by applying non-individualized head-related transfer functions (HRTFs), downloaded from an MIT website (sound.media.mit.edu), to the stimuli. In one condition, the background speakers were colocated with the apparent target location, corresponding to 0° azimuth. In another condition, the target speech was at a 0° azimuth, one background speaker was at +15° and the other at −15° azimuth. To make the task more challenging, the speech stimuli were also convolved with the impulse response of a reverberant room (T₆₀ = 1.07s) after HRTFs were applied. The four conditions, 2 target/maskers F0 (same/different) configurations x 2 speaker locations (colocated/non-colocated) were presented in a randomized order across participants. In all conditions, the target speech was presented at 75 dB SPL. In the same-F0 conditions, the signal-to-noise ratio (SNR) was 2 dB, while in the different-F0 conditions, the SNR was lowered to 0 dB to prevent ceiling performance. For participants who had high-frequency hearing thresholds above 20 dB HL, the stimuli were amplified in the region of hearing loss using a half-gain rule. The amplification was implemented by filtering the speech stimuli into one-third octave bands using overlapping (at cutoff frequencies) 4th order Butterworth filters and Matlab’s filtfilt function. The full spectrum speech was reconstructed by summing the outputs of the filters after the gain of high-frequency filters was adjusted according to the participant’s hearing-loss configuration. For filters with center frequencies between audiometric octave frequencies, the amounts of gain were interpolated from the gain values based on the hearing loss at the nearest octave frequencies. The same processing was performed for speech presented to normal-hearing participants but without applying gain.

The four conditions were repeated using speech stimuli from the same corpus that were lowpass filtered to limit their spectra to the frequency range over which all the participants had normal hearing sensitivity by clinical standards. A 4th-order Butterworth lowpass filter with a cutoff frequency of 2 kHz was used to limit the spectrum. The filtering was performed using Matlab’s filtfilt function to remove phase shifts across frequency. Spread of excitation into high-frequency regions of the cochlea was masked by presenting the lowpass-filtered speech with a highpass-filtered Gaussian noise. A 4th-order Butterworth filter was used to create the masking noise with a bandwidth of 2.2 – 10 kHz. The level of the noise was 10 dB below that of the target speech. For the lowpass-filtered speech, the SNR (target-to-babble ratio) was set to 6 dB in the same-F0 conditions and to 4 dB in the different-F0 conditions.

The two-talker masker was created by concatenating sentences spoken by the maskers from lists that were not used for the target speech in any test conditions for a given participant. The masker waveforms were summed and segmented so that the masking speech was longer than the target by at least 1 second (the masker duration was set relative to the longest sentence in the corpus). In each trial, the target started after a 0.8 sec delay from the onset of the two-talker babble. The highpass noise in the lowpass-speech condition was 2 sec longer than the two-talker masker. The two-talker masker with the target started 0.8 sec after the noise onset.

The experiment began with practice during which one list was presented for each test condition using the full speech spectrum. None of the lists used during the practice was repeated during the testing. Two lists of 10 sentences (a total of 20 sentences) were used for each test condition. After each sentence, participants were prompted to type the sentence they heard. If they missed parts of the sentence, they were asked to type any of the words they heard and were encouraged to make their best guess even if they were not sure about the exact words. No feedback was provided during training or data collection. Five keywords in each sentence were used to obtain a score. Words that were misspelled (rather than not heard) were counted as correct responses. The proportion of correctly recalled keywords across two lists was used as the final score for each test condition.

During testing, participants were seated in a single-walled sound-attenuating booth. The stimuli were played using Matlab (Mathworks, Natick, MA) via an E22 soundcard (LynxStudio, Costa Mesa, CA) with a sampling rate of 44.1 kHz and were delivered to both ears via Sennheiser HD650 headphones (Sennheiser, Old Lyme, CT).

4.3. Results and discussion

The proportion correct scores from each condition were converted to rationalized arcsine units (RAUs; Studebaker 1985). Fig. 6 shows the mean RAU scores for the same- and different-F0 conditions in the upper and lower rows, respectively. The left column shows scores for the full-spectrum speech and the right column shows scores for the lowpass-filtered speech. The scores for the corresponding colocated and non-colocated conditions (within each panel) were obtained using the same SNRs, and the scores in different panels were obtained using different SNRs, as described in Section 4.2, and thus cannot be directly compared.

Fig. 6.

Speech-recognition scores (in RAUs) for target speech in two-talker babble, in the colocated (target/maskers at 0⁰ azimuth) and non-colocated (target at 0⁰ and maskers at ±15⁰ azimuths) conditions. The error bars indicate ± 1 SE of the mean. The upper and lower panels show data for the target and masker speakers with the same and different target/maskers voice-F0s, respectively. The left and right columns show data for the full-spectrum and the lowpass-filtered speech, respectively. Asterisks denote significant differences between age groups (*** p < 0.001, ** p < 0.01, and * p < 0.05).

Four mixed-design ANOVAs were performed on the RAU scores separately for the four conditions (2 F0 × 2 speech spectrum), with relative spatial location of the target and maskers (colocated/non-colocated conditions) as a within-subjects factor, age group as a between-subjects factor, and average hearing threshold as a covariate. For the full-spectrum speech, the average hearing threshold was calculated across frequencies from 0.25 to 8 kHz (ALLAVG), because there were significant differences in hearing thresholds between the three age groups across the entire audiometric range (see Fig. 1). For the lowpass-filtered speech, the average hearing threshold was calculated for octave frequencies from 0.25 to 2 kHz (LFAVG), because speech information was limited to this frequency range.

The ANOVA for the same-F0 full-spectrum condition showed significant effects of spatial condition [F(1, 54) = 14.17, p < 0.001], age group [F(2, 54) = 5.34, p = 0.008], and ALLAVG [F(1, 54) = 4.67, p = 0.035]. There was no significant interaction between spatial condition and age group [F(2, 54) = 1.21, p = 0.307] or between spatial condition and ALLAVG [F(1, 54) = 1.06, p = 0.308]. Post-hoc pairwise comparisons showed that in the colocated condition, the youngest group performed significantly better than the middle group (p = 0.003) and the oldest group (p <0.001), but there was no significant difference in scores between the middle and oldest groups (p > 0.05). Performance in the non-colocated condition declined progressively with increasing age (youngest vs. middle, p = 0.006; youngest vs. oldest, p < 0.001; middle vs. oldest, p = 0.04).

The ANOVA for the different-F0 condition showed no significant difference between the two spatial conditions [F(1, 54) = 3.34, p = 0.073]. There was a significant effect of age group [F(2, 54) = 6.06, p = 0.004], but no significant interaction between spatial condition and age group [F(2, 54) = 1.77, p = 0.181] indicating overall worse performance of older than younger individuals in both colocated and non-colocated conditions. Post-hoc pairwise comparisons for the colocated condition showed significant differences between the youngest and middle groups (p = 0.001) and the youngest and oldest groups (p < 0.001), but not the middle and oldest groups (p > 0.05). Similarly, in the non-colocated condition, there was a significant difference in performance between the youngest and middle (p = 0.006) and youngest and oldest (p < 0.001), but not middle and oldest (p > 0.05) groups. The ANOVA showed no significant effect of ALLAVG [F(1, 54) = 2.93, p = 0.093] and no interaction between spatial condition and ALLAVG [F(1, 54) = 0.10, p = 0.752].

The ANOVA for the same-F0 lowpass-filtered condition showed no significant effect of spatial condition [F(1, 54) = 0.41, p = 0.526], indicating that spatial cues were insufficient to improve speech recognition. There was a significant effect of age group [F(2, 54) = 3.48, p = 0.038], but no significant interaction between spatial condition and age group [F(2, 54) = 0.68, p = 0.513]. Post hoc pairwise comparisons showed that in the colocated condition, the youngest group performed significantly better than the middle (p = 0.021) and the oldest (p = 0.01) groups, but there was no significant difference in performance between the middle and oldest groups (p > 0.05). In the non-colocated condition, there was a significant difference in performance between the youngest and oldest groups (p = 0.01) but not between the youngest and middle or middle and oldest groups (both p > 0.05). For the same-F0 lowpass-filtered speech, the ANOVA showed a significant effect of LFAVG [F(1, 54) = 5.67, p = 0.021], but no interaction between spatial condition and LFAVG [F(1, 54) = 0.47, p = 0.497].

For the different-F0 lowpass-filtered speech condition, the ANOVA showed no significant effect of spatial condition [F(1, 54) = 1.80, p = 0.185] but there was a significant effect of age group [F(2, 54) = 4.54, p = 0.015] and a significant interaction between condition and age group [F(2, 54) = 3.69, p = 0.032]. Post hoc pairwise comparisons revealed that for the colocated condition there was a significant difference in performance between the youngest and oldest groups (p = 0.041) but not between the youngest and middle or middle and oldest groups (both p > 0.05). In the non-colocated condition, the youngest group performed significantly better than both the middle (p = 0.037) and oldest (p = 0.001) groups but there was no significant difference between the middle and oldest groups (p > 0.05). LFAVG was not a significant factor [F(1, 54) = 0.58, p = 0.451] and there was no significant interaction between LFAVG and spatial condition for the low-pass filtered speech with different target/masker F0s [F(1, 54) = 1.85, p =0.18].

Overall, the pairwise comparisons showed that for the full-spectrum speech, a significant decline in speech perception was observed for the middle-aged compared to the young adults and this was true when the target and maskers were colocated or non-colocated. The age effects on the perception of speech in the colocated condition in this study appear inconsistent with the lack of significant effects of age in speech-on-speech masking measured using diotic presentation in the study of Prendergast et al. (2019). Differences in experimental procedures may have contributed to the different outcomes. Prendergast et al. (2019) used a closed-set speech corpus (coordinate response measure) for both the target and the two-talker masker while IEEE sentences were used in this study. Another difference was that Prendergast et al. (2019) measured the SNRs needed for performance corresponding to 25% correct word recognition. In this study, SNR = 0 dB was used in the comparable (diotically presented full-spectrum speech) condition. For the low-pass filtered speech, our finding of significant age effects on performance in the colocated condition is consistent with that reported by Léger et al. (2014) for perception of vowel-consonant-vowel syllables in the presence of a single interfering talker.

The RAU scores were used to calculate SRM separately for each voice-F0 and speech-spectrum condition by subtracting the score for the colocated target /maskers from the score for the non-colocated target/maskers. The magnitudes of SRM for the full-spectrum and the lowpass-filtered speech are shown in the left and right panels of Fig. 7, respectively. In each plot, the left three bars show SRM for the same-F0, and the right three bars show SRM, for the different-F0 conditions.

Fig. 7.

Spatial release from masking, for the three age groups, for full-spectrum speech (left panel) and for lowpass-filtered speech (right panel). The left and right set of bars show SRM for the same and different target/masker voice-F0s, respectively. The error bars indicate ± 1 SE of the mean. Asterisks denote significant differences (* p < 0.05).

Although age group was a significant factor for speech understanding in all four conditions, the SRM, which is a differential measure reflecting the benefit of spatial segregation, was not always affected by age. Four one-way ANOVAs were run, each for one F0 and speech-spectrum combination. A significant effect of age was observed for the same-F0 full spectrum condition [F(2, 57) = 5.93, p = 0005] and the different-F0 full-spectrum condition [F(2, 57) = 7.8, p = 0.001]. Post hoc pairwise comparisons showed significant differences between the youngest and oldest groups (p = 0.006 for the same-F0 condition, and p = 0.001 for the different-F0 condition). No other pairwise comparisons were significant (all p > 0.05). No significant age effects were observed for the lowpass-filtered speech [F(2, 57) = 0.58, p = 0.56 for the same-F0 condition, and F(2, 57) = 2.74, p = 0.07 for the different-F0 condition].

Treating age as a continuous variable, correlations between the SRM and age, and average hearing threshold, were performed separately for each voice-F0 and speech-spectrum conditions with participants pooled across the three age groups. The scatterplots in Fig. 8 show the individual magnitudes of SRM for the full-spectrum speech (left panel) and the lowpass-filtered speech (right panel). In both panels, the grey and red circles show the data for the same- and different-F0 conditions, respectively.

Fig. 8.

Scatterplots in the left and right panels show individual SRM plotted as a function of age for the full-spectrum and lowpass-filtered speech, respectively. Grey and red symbols show data for the same- and different-F0 conditions, respectively. The lines represent a linear regression fit to the data (Pearson product correlations coefficients are shown in Table 4).

The results of the correlational analyses are shown in Table 4. For the same-F0 full-spectrum speech, SRM was significantly correlated with age and ALLAVG (p < 0.001 for age, and p =0.003 for ALLAVG) at the Bonferroni-corrected significance level (∝ = 0.006 for eight comparisons). After controlling for one variable, the effect of age was significant only at the uncorrected level (p = 0.022) while ALLVG was not (p > 0.05). For the different-F0 full-spectrum speech, only the correlation between SRM and age was significant (p < 0.001) and it remained significant after partialling out ALLAVG (p = 0.003) at the corrected significance level. For the lowpass-filtered speech, neither age nor LFAVG were significantly correlated with SRM (all p > 0.05).

Table 4.

Coefficients of Pearson’s product correlations between spatial release from masking (SRM) and age, and between SRM and pure-tone average (PTA) threshold. For the full-spectrum speech, PTA was equal to ALLAVG and for lowpass-filtered speech, PTA was calculated across frequency range 0.25 – 2 kHz. For partial correlations, the variable controlled for is listed in parentheses. Bonferroni-corrected significant correlations are shown in bold font (p < 0.025). Bonferroni correction was applied separately for the full-spectrum and lowpass speech conditions.

SRM	Age	PTA	Age (PTA)	PTA (Age)
Same F0 (full spectrum)	−0.47	−0.38	−0.30	−0.07
Same F0 (lowpass)	−0.07	0.07	−0.11	0.11
Different F0 (full spectrum)	−0.48	−0.32	−0.38	0.05
Different F0 (lowpass)	−0.26	0.02	−0.25	0.17

As shown by the scatterplot in the right panel of Fig. 8, the SRM magnitudes for the lowpass-filtered speech were distributed around zero, indicating that lowpass filtering eliminated the benefit of spatial segregation. It is worth noting that even for the full-spectrum speech, a number of participants across the adult lifespan did not exhibit a positive SRM (see the points at and below the dashed line in the left panel of Fig. 8).

Consistent with our findings for the full-spectrum speech, Gallun et al. (2013) also reported a significant effect of age, independent of hearing loss, on SRM when using the same spatial locations for the target and maskers as the ones used here. In contrast, audiometrically-matched groups of young and older adults in the study of Füllgrabe et al. (2015) did not differ in SRM, but only relatively large target\maskers spatial separations (± 60°) were used in that study. Several other studies, which used young and older groups (with not well-matched normal and near-normal audiometric thresholds), also reported no age-related differences in the ability to use spatial cues to improve speech recognition (e.g., Cameron et al. 2011; Gelfand et al. 1988; Li et al. 2004; Singh et al. 2008). All these studies used larger target\maskers spatial separations (the smallest spatial separation was 45°) than the ±15° used in the present study. A few studies found a significant effect of hearing sensitivity but not age on SRM despite applying amplification to spectral regions corresponding to hearing loss (Glyde et al. 2015; Glyde et al. 2013). Srinivasan et al. (2016) showed that for individuals with age-related hearing loss, age was a dominant predictor of SRM for small target\maskers spatial separations while the degree of hearing loss dominated SRM for large spatial separations. This could explain the differences in findings between this and the Gallun et al. (2013) study versus those reported by Glyde et al. (2013), which used maskers located at ±90°. An apparent problem with this explanation is that Srinivasan et al. (2016) found hearing loss to be a dominant predictor of SRM for target\maskers spatial separations above ±8°, which includes the ±15° separations used in this study. However, participants in the Srinivasan et al. (2016) study had substantially greater loss of hearing sensitivity than the middle and oldest participants tested here. In addition, reverberation was added to our speech stimuli. It is possible that these differences contributed to our finding that age rather than hearing sensitivity significantly influenced SRM for target/masker spatial separations of ±15°.

As shown in Fig. 7, lowpass filtering of the stimuli reduced the amount of SRM for all age groups, indicating significant contributions of frequency components above 2 kHz to SRM. This is consistent with data from Kidd et al. (2010) showing that lowpass filtering reduces SRM and that significant SRM is observed for speech that is bandpass filtered into a range between 3 and 6 kHz. It is possible that age effects on SRM were not seen for the lowpass-filtered speech because of a very small or absent SRM for many participants.

5. Cognitive assessment

5.1. Stimuli and Procedure

All participants who completed the speech task also completed a cognitive trail making test (TMT; Reitan 1955) implemented in the Psychology Experiment Building Language platform (PEBL version 2.0.4; Mueller and Piper 2014). The test consists of two parts, A and B. In part A, participants were presented with a display of encircled numbers from 1 to 25 that were distributed out of order on a computer screen. The task was to click on the numbers (using a computer mouse) in ascending order. In task B, the display on the screen contained randomly distributed numbers from 1 to 12 and letters of the alphabet from A to L. Participants were asked to click on the numbers and letters in alternating sequence in ascending order, i.e., 1-A-2-B-3-C-, etc. For both parts, participants were instructed to complete the task as fast and accurately as possible. The time needed to complete each test was recorded and used as a score. Scores from part A are thought to be influenced by such cognitive functions as processing speed, psychomotor control, and visual search, whereas scores from part B additionally index executive function (Bowie and Harvey 2006). The TMT test was selected because previous studies have shown that performance on this test accounted for significant amounts of the variance in just detectable interaural time differences (Füllgrabe et al. 2015; Shehorn et al. 2020; Strelcyk et al. 2019) and the variance in speech scores in a speech-on-speech task involving spatially separated targets and maskers (Füllgrabe et al. 2015; Woods et al. 2013). Prior to test administration, participants completed shorter practice versions of each part.

5.2. Results and discussion

The mean times needed to complete the TMT-A test were, 24.3 s (SE = 1.5) for the youngest group, 27.2 s (SE = 1.5) s for the middle group, and 30.2 s (SE = 1.5) s for the oldest group. For the TMT-B test, the mean times were, 28 s (SE = 1.7) for the youngest, 34.2 s (SE = 2.6) for the middle, and 37.3 s (SE = 2.3), for the oldest group. A repeated-measures ANOVA performed on scores from the two parts of TMT test, with TMT task (A and B) as a within-subjects factor and age group as a between-subjects factor, showed significant effects of TMT task [F(1,54) = 50.58, p < 0.001] and age group [F(2,54) = 5.828, p = 0.005], but no significant interaction [F(2,54) = 2.51, p = 0.09]. Post hoc analyses showed that the oldest group took significantly longer to complete the tests than the youngest group (p = 0.018), but there were no significant differences between the scores for the oldest and middle, and the middle and youngest groups (both p > 0.05). As expected, participants took significantly longer to complete part B than part A of the test (p < 0.001). For all participants combined, there were significant correlations between TMT scores and age (part A: r = 0.33, p = 0.012; and part B: r = 0.32, p = 0.017) suggesting some decline in cognitive abilities of our participants with increasing age. Because the difference between scores from parts B and A is thought to reflect executive function (Sánchez-Cubillo et al. 2009), a one-way ANOVA was performed using this differential measure. There was a significant effect of age group [F(2, 57) = 8.77, p < 0.001]. A post hoc analysis showed that the difference scores were significantly larger for the oldest than the youngest group (p < 0.001) and for the middle than the youngest group (p = 0.016), but there was no significant difference between the middle and oldest groups (p = 0.056). Overall, these results suggest that a decline in executive function emerges in middle age (in the 40s). There was a significant correlation between the differential TMT(B-A) measure and age treated as a continuous variable (r = 0.392, p = 0.002).

6. Correlation and regression analyses

Data from the 27 participants who completed all of the experiments in this study were used to investigate the strength of associations between the obtained measures using multiple correlations. Figure 9 shows a heat map representing values of the correlation coefficients. Since the correlations were exploratory in nature, the significant effects indicated by white asterisks are based on p values that were compared against the uncorrected significance level (α = 0.05). As indicated by the darker areas with asterisks near the diagonal, measures from the same experiment but for the different parameters were significantly correlated with one another while measures from different tasks were uncorrelated in most cases, even though, with the exception of the cognitive tests, the measures were expected to tap into the same general mechanisms that underlie temporal-envelope processing. There were a few weak correlations between measures from different experiments. The scores from both parts of the TMT test (correlation not included in the heat map) and the differential TMT(B-A) scores were negatively correlated with the EFR for one condition (0-dB modulation depth in quiet) and positively correlated with envITD thresholds for a 2-kHz tone with 100-Hz AM in quiet. All envITD thresholds were significantly correlated with SRM for the different-F0 low-pass-filtered speech condition, but for the different-F0 full-spectrum speech, SRM was significantly correlated only with envITD thresholds for the 2-kHz AM tones presented in quiet. For the same-F0 full-spectrum speech, there was a significant correlation between the SRM and envITD thresholds for the 2-kHz carrier but only for 100-Hz AM, while for the same-F0 lowpass speech, SRM was correlated with the 2-kHz envITD thresholds but only for the 40-Hz AM. Additionally, SRM for the same-F0 lowpass speech was correlated with gap-detection thresholds at 4 kHz.

Fig. 9.

A heat map showing associations between the measures collected in this study. Asterisks denote significant correlations (*** p < 0.001, ** p < 0.01, and * p < 0.05) at the uncorrected significance level.

A multivariate linear regression model was used to assess the proportion of variance in SRM explained separately and jointly by three variables: age, envITD for the 2-kHz carrier (averaged for the two modulation rates, 40 and 100 Hz), and ALLAVG. No other measures were entered into the regression model since they were not significantly correlated with SRM. Only the full-spectrum conditions were considered in the regression analysis because little or no SRM was observed for the lowpass-filtered speech for many participants. First, associations between each of the independent variables were tested using simple linear regressions. There was a significant correlation between age and ALLAVG (r = 0.48, p = 0.011), but the collinearity between the two variables was relatively weak so both variables were entered in the regression model. The average envITD thresholds for the 2-kHz carrier were not significantly correlated with age (r = 0.25, p = 0.21) or with ALLAVG (r = 0.22, p = 0.28).

For the same-F0 condition, all three independent variables were significant predictors of SRM, with age explaining 25% of the variance [F(1, 26) = 9.52, p = 0.005], envITD for the 2-kHz carrier explaining 16% of the variance [F(1, 26) = 5.97, p = 0.022], and ALLAVG explaining 14.5% of the variance [F(1, 26) = 5.43, p = 0.028], when considered separately. Jointly, the three variables explained 31% of the variance in SRM. With all three variables entered into the model, age was the dominant (and the only significant) predictor (25% of the variance explained), with envITD and ALLAVG adding only nonsignificant 5 and 1.4 percentage points to the variance explained, respectively.

For the different-F0 condition, only two variables, age and average envITD threshold, were significant predictors of SRM. Age explained 42% of the variance [F(1, 26) = 20.08, p < 0.001] and average envITD threshold explained 15% of the variance [F(1, 26) = 5.61, p = 0.026] when considered separately. Jointly, the two variables explained 48% of the variance in SRM [F(2, 26) = 12.95, p < 0.001], with age being the only significant predictor (42%) and average envITD threshold adding only nonsignificant 6 percentage points to the variance explained.

In summary, although the proportions of variance in SRM explained by the independent variables in both linear regression models were small, age emerged as the main predictor. The cognitive abilities evaluated in this study using two TMT tests declined with increasing age, but this decline did not explain the age-related reduction in SRM in any of the speech conditions since neither the scores from TMT parts A and B nor the difference between them were significantly correlated with SRM (only the correlation with the difference score is shown in Fig. 9).

7. General discussion

7.1. Sensitivity of temporal-processing measures to cochlear synaptopathy

A common hypothesis regarding perceptual consequences of cochlear synaptopathy is that the diffuse loss of synapses should lead to degraded temporal processing of medium-to-high-level sounds with fluctuating temporal envelopes, particularly in noisy backgrounds (e.g., Bharadwaj et al. 2015; Bharadwaj et al. 2014; Chambers et al. 2016; Kujawa and Liberman 2009; 2015; Moore et al. 2019; Plack et al. 2014; Schmiedt et al. 1996). None of the experiments performed in this study are new, but previous studies have not always used stimuli and conditions for which effects of cochlear synaptopathy are most likely to be revealed and\or examined the associations between the different measures. The goal of this study was to test the same group of participants in tasks that rely on temporal-envelope processing using stimuli that promote contributions from medium- and low-spontaneous-rate fibers, which are most vulnerable to cochlear synaptopathy (Furman et al. 2013).

In humans, direct evidence of cochlear synaptopathy from temporal bone studies shows that loss of peripheral synapses increases with advancing age (Makary et al. 2011; Viana et al. 2015; Wu et al. 2019). Given this evidence, it was hypothesized that all the measures of temporal envelope processing collected in this study should be adversely affected by age, and thus should be significantly correlated with age and each other. However, except for a few seemingly random cases (gap detection for a 4-kHz marker, and EFRs for 0-dB AM in quiet and −8 dB AM in noise), no significant age effects on behavioral and neural (EFR) measures were found. For a subset of 27 participants who completed all the tests, no correlations were found between the different measures (see Fig. 9). Given the direct evidence for age-related cochlear synaptopathy in humans (Wu et al. 2019), the likely explanation for the lack of significant correlations is that the measures used in this study are simply not sensitive to cochlear synaptopathy. The broader implication of this finding is that the lack of test sensitivity rather than the lack of noise-induced cochlear synaptopathy in humans could be the reason why previous studies using similar measures failed to show significant correlations with noise exposure (e.g., Grose et al. 2017; Guest et al. 2017; Prendergast et al. 2017a; Prendergast et al. 2017b; Yeend et al. 2017). Our findings are consistent with another recent study on age-related cochlear synaptopathy that used similar, although not identical, envelope-based measures (Prendergast et al. 2019).

Although we found significant effects of age on SRM, none of the measures of temporal-envelope processing significantly contributed to the variance in SRM. Given the likely insensitivity of our behavioral and EFR-based measures to cochlear synaptopathy, it is impossible to make inferences about the extent to which the reduction of SRM with age can be attributed to peripheral synaptic loss. Hearing sensitivity and a non-auditory (TMT) measure of executive function were not significant predictors of SRM. However, increasing age is known to be associated with changes in neural coding in midbrain and cortical areas (Alain 2014; Anderson et al. 2012; Presacco et al. 2016; Tremblay et al. 2003), and a decline in other cognitive functions that were not tested here (Craik and Salthouse 2011; Füllgrabe et al. 2015; Humes et al. 2013; Van der Linden et al. 1994; Ward et al. 2016). These central changes may have affected both speech-recognition performance and SRM. However, some of the changes in neural processing at more central sites may occur as a consequence of both hair-cell damage and cochlear synaptopathy rather than originating from deficits at those sites, as discussed below.

7.2. Compensatory neural gain in midbrain and cortical responses

The lack of sensitivity of the behavioral and electrophysiological measures used in this study to cochlear synaptopathy may result from plastic neural changes at central sites in response to degraded peripheral input (Chambers et al. 2016; Henry et al. 2017; Parthasarathy et al. 2019; Parthasarathy and Kujawa 2018). These plastic changes occur due to a disrupted balance between excitatory and inhibitory neural inputs along the auditory pathway (Caspary et al. 2008; Walton 2010). A compensatory increase in synchronized neural activity in response to reduced peripheral input can be gleaned from EFRs measured for a wide range of modulation rates in animals with noise-, age-, or ouabain-induced cochlear synaptopathy (Parthasarathy and Kujawa 2018; Shaheen et al. 2015). Although the neural generators of EFRs cannot be localized directly, their generation sites can be inferred based on decreasing limits of phase locking along the ascending auditory pathway (Joris et al. 2004). EFRs measured in animals with cochlear synaptopathy are reduced compared with those measured in controls for high (> 600–800 Hz) but not for lower modulation rates (Parthasarathy and Kujawa 2018; Shaheen et al. 2015). Similarly, aging has an adverse effect on EFRs but only for high modulation rates (Parthasarathy and Kujawa 2018), suggesting that rates that elicit responses synchronized to the stimulus envelope at the level of the midbrain and cortex are relatively enhanced. Parthasarathy et al. (2019) used both EFR measures and invasive multiunit recordings from IC neurons in large samples of younger and older rats. They found that neural activity synchronized to the envelope and the fine structure of a speech-like stimulus was reduced at the input to IC neurons in older animals, but the output synchronized rate from the midbrain was relatively more enhanced in the older group. An even stronger increase in synchronized responses occurs in the cortex (Chambers et al. 2016). Consequently, responses to lower modulation rates at central sites are relatively unaffected by degraded peripheral input while higher modulation rates may not benefit from a similar compensatory enhancement (Parthasarathy and Kujawa 2018).

In humans with normal hearing, the temporal modulation transfer functions measured for high-frequency tonal carriers (Kohlrausch et al. 2000), show greater sensitivity to modulation rates that elicit synchronized neural responses at more central sites <~100–150 Hz for tonal carriers) than to higher modulation rates that are represented by a (non-synchronized) rate code (e.g., Wang et al. 2008). This indicates that the rate code is less precise for coding temporal fluctuations. It may be that modulation rates coded by rate rather than synchrony to the envelopes are not subject to the same neural gain as the synchronized responses, which could lead to perceptual effects of cochlear synaptopathy for high but not lower modulation rates. In humans, older adults exhibit reduced EFR magnitudes for modulation rates above, but not below, ~100–150 Hz (Purcell et al. 2004) compared to young adults. Based on animal studies, this pattern of results is consistent with an age-related reduction in temporal envelope coding at peripheral sites that, for rates lower than ~100–150 Hz, is offset by a compensatory gain in neural synchronization at central sites. Taken together, these findings may provide an explanation for the lack of significant age effects in the present study, as with the exception of AM-detection for a noise carrier, modulation rates ≤ 100 Hz were used. It may also explain why significant age effects were observed for gap-detection thresholds for the 4-kHz carrier. The duration of gaps at threshold approximated one cycle of AM with a rate > 100 Hz in all three age groups (see Fig. 3). Thus, it may be that the modulation rates used in our envITD and EFR experiments were too low to reveal effects of age-related cochlear synaptopathy.

This explanation for the lack of significant findings has some weaknesses. First, it is unclear whether the compensatory neural gain at central sites is limited to modulation rates represented by neural activity that is synchronized to the envelope fluctuations. Second, although neural enhancements have been shown for AM tones and complex stimuli presented in quiet, it has been suggested that the coding precision for stimuli presented in noise remains degraded even in cortical responses (Chambers et al. 2016; Parthasarathy and Kujawa 2018). In this study, significant age effects on EFRs were only observed for fully (100%) modulated tones presented in quiet, a condition in which the enhanced synchronized responses would be expected to be most effective at compensating for the reduced peripheral input due to age-related cochlear synaptopathy.

Some (indirect) support for the modulation-rate-based explanation of the failure to observe patterns consistent with cochlear synaptopathy comes from a recent study of Mepani et al. (2021). EFRs were measured for 120-Hz rectangular modulation with a relatively short (25%) duty cycle. Mepani et al. (2021) found that the high-rate EFR components (360 – 600 Hz) but not the lower-rate (120–240 Hz) components were larger for participants who performed better on various speech-recognition tasks. The correlations remained significant after controlling for audiometric threshold leading to the conclusion that the results reflected effects of cochlear synaptopathy. It should be noted that the EFRs were measured for stimuli in quiet, and thus it is unclear if similar patterns of correlations would be observed for tones with rectangular AM presented in noise. With many questions still unresolved, conclusions about the reasons for the insensitivity of our measures to cochlear synaptopathy remain speculative.

7.3. Other limitations

Although our initial recruitment criteria required all participants to have normal (< 20 dB HL) hearing thresholds at all audiometric frequencies (0.25 – 8 kHz), after screening and turning away a large number of middle-aged and older volunteers with mild high-frequency hearing loss, we decided to include those with hearing loss ≤ 40 dB HL limited to frequencies ≥ 4 kHz to reach the targeted sample sizes for the two older groups. Because not all middle and older participants had hearing loss, the correlation of hearing loss with age was relatively weak (r = 0.48). The approach was to statistically partial out effects that were due to hearing loss. This approach was not optimal for a few reasons. First, cochlear hair-cell damage and synaptopathy likely coexist and thus by partialling out effects of hearing loss, some portion of the variance in our measures that was due to cochlear synaptopathy was likely removed (Martin et al. 1991). Second, cochlear hair-cell damage causes plastic changes at central sites of the auditory system (Chen et al. 2013; Ison and Allen 2003; Sun et al. 2009). Although we cannot rule out the possibility that such changes affected the data for some participants in the middle and oldest groups, given that only mild hearing loss was present, the plastic changes probably did not play an important role. This is supported by the non-significant age group differences in most of our experiments.

An age-related decrease in overall speech performance and in SRM in speech-on-speech masking is often considered a likely consequence of cochlear synaptopathy. Significant age effects on SRM were observed after controlling for differences in hearing thresholds. The effects were not accounted for by age-related changes in cognitive measures of processing speed and executive function. However, speech performance can be affected by other cognitive functions such as selective attention (Best et al. 2007; Best et al. 2008; Clayton et al. 2016; Holmes and Griffiths 2019) and working memory (Akeroyd 2008; Clayton et al. 2016). These cognitive factors may be affected by age, independent of peripheral hearing status. Increased cognitive demands in challenging speech conditions leads to increased listening effort (Winn et al. 2015; Zekveld et al. 2014) that may vary across participants depending on motivation (Pichora-Fuller et al. 2016). Although age-related reductions in speech performance and SRM are an expected consequence of cochlear synaptopathy, limited testing for cognitive function, and the lack of other measures sensitive to age-related cochlear synaptopathy in this study, makes it impossible to attribute these reductions to synaptopathy.

Finally, the premise for all the experiments in this study was that our stimuli were mainly encoded by medium- and low-spontaneous-rate fibers that are most vulnerable to synaptopathy (Furman et al. 2013). However, Carney (2018) questioned the role of medium- and low-spontaneous-rate fibers in the coding of temporal envelope fluctuations for medium and high-level stimuli. Carney (2018) used model simulations to show that robust temporal information may be available in neural fluctuation profiles of high-spontaneous-rate fibers even for stimulus levels for which their average firing rates are saturated. This alternative view does not necessarily mean that cochlear synaptopathy would have no consequences for perception of complex sounds in noise. Encina-Llamas et al. (2019) used a computational model to predict the growth of EFR magnitude with increasing level of an AM tone that was measured in a small number of listeners with normal hearing and with mild high-frequency hearing loss. Predictions of their EFR data were improved when some loss of all three types of AN fibers was simulated compared with predictions simulated assuming loss of only medium- and low-spontaneous rate fibers. However, it is likely that if the modulation-profile coding by high-spontaneous-rate fibers were dominant even for masked high-level stimuli, cochlear synaptopathy could have lesser consequence on perception as high-spontaneous-rate fibers are less affected by synaptopathy (Furman et al. 2013). The model simulations in the study of Carney (2018) did not include medium- and low-spontaneous-rate fibers. It is possible that these fibers would increase the precision of coding of envelope fluctuations (Joris and Yin 1992) and the contrast between the modulation profiles across frequency, that are at the center of the complex-sound coding strategy proposed by Carney.

8. Conclusions

Despite the direct evidence of age-related cochlear synaptopathy in humans (e.g., Wu et al. 2019), we found no effects of age on behavioral and EEG-based measures that have been used in other studies to search for consequences of cochlear synaptopathy (Bharadwaj et al. 2014; Prendergast et al. 2019; Prendergast et al. 2017a). The lack of significant effects suggests that these measures are insensitive to synaptopathy, and that they are inadequate for studying effects of age or effects of noise-induced cochlear synaptopathy.

Highlights:

• Psychophysical and electrophysiological measures of temporal envelope processing are not correlated with each other for listeners with normal and near-normal hearing.

• Gap detection thresholds for tones in noise worsen with advancing age.

• Age, independent of hearing loss, can substantially reduce spatial release from masking for relatively small azimuthal separations.

• Cognitive executive function declines with increasing age, but this decline does not predict the reduction in spatial release from masking for listeners aged 20 – 66 years.