Arousal state fluctuations are a source of internal noise underlying age-related declines in speech intelligibility

Abstract

Understanding speech in noisy, multi-talker environments is crucial for social communication but becomes increasingly challenging and frustrating as we age. Here, we simulated the acoustic challenges of multi-talker listening and found that adults over 50 years old (N = 76) recognized speech more slowly, less accurately, and less consistently than younger adults (N = 107). While peripheral hearing status accounted for average differences in speech intelligibility by age, it did not account for moment-to-moment variability in speed and accuracy - fluctuations central to the frustration experienced by older listeners in challenging environments. We hypothesized that age-related changes in brain arousal systems might account for the fluctuant "noise" in speech processing observed in older listeners. To isolate the contribution of arousal state independent of hearing status and cognitive load, we measured the pre-stimulus pupil-indexed arousal state (PPAS) immediately prior to speech onset. Older - but not younger - adults exhibited a striking inverted-U relationship between PPAS and speech recognition accuracy. Notably, pupil-indexed listening effort measured seconds later during speech encoding was not associated with trial-to-trial performance. Moreover, older adults exhibited altered arousal regulation, occupying a lower PPAS extremum not observed in younger listeners that was specifically associated with performance deficits and subjective listening difficulties reported in hearing health questionnaires. These findings show that age-related changes in central arousal states interact with peripheral hearing status to offer a more complete explanation for why older adults find speech processing in social setting so challenging.

Figure 1.

Performance is impaired at low and high levels of pupil-linked arousal in older adults. (A) Schematic of the multi-talker digit recognition task in which study participants were asked to report four digits vocalized by a target speaker simultaneously masked by two additional distracting speakers (Dstrct1, Dstrct2). (B) Outcomes on individual 0 dB SNR trials for two study participants, highlighting the presence of trial-to-trial variance. The red line shows mean performance and the grey shaded region the standard deviation. (C) An example pupil trace recorded during a 0 dB SNR block. Trial numbers are shown along the abscissa and pre-stimulus window are underlaid in pink. (D) A histogram of study participants split at age 50 into younger and older adult cohorts. (E) Older adults (orange square) correctly reported a significantly smaller proportion of digits in comparison to younger adults (blue circle) during 0 dB SNR trials (two-sample t-test, P<0.001, N=107/76). This trend generalized for age cut-offs above and below 50. Error bars/shaded region denote standard error. (F) Dependence of task accuracy on pre-stimulus pupil-indexed arousal state (PPAS) during 0 dB SNR trials. To normalize for variations in pupil size across individuals and age, PPAS values were assigned to one of five equally sized bins per person. Smaller and larger diameters are associated with increased errors in older adults. (G) We measured differences in accuracy between the 3^rd and 1^st/5^th quintiles (Δ at extremes). Accuracy at pupil-indexed extremes significantly decreased in the older cohort (one-sample t-test, P<0.001, N=76), and were again robust to the exact choice of age cut-off. (H-J) as (E-G) for response times (RT). Older adults took significantly longer to respond than younger adults (two-sample t-test, P<0.001, N=107/76) with increased response times at pupil-indexed extremes during 0 dB SNR trials (one-sample t-test, P=0.002, N=76).

Figure 2.

Older adults exhibit a broader range of pupil-indexed arousal. (A) Stacked plots show pupil quintile likelihoods as a variable of trial number (8 trial moving average). Younger and older adults are shown in blue and orange, respectively, with smaller pupil sizes indicated by quintiles of a darker shade. (B) The proportion of trials spent in the smallest pupil quintile (Q1) increased between the start and the end of the experiment for younger adults (paired t-test, P<0.001, N=106), but not older (paired t-test, P=0.68, N=73). (C) State transition statistics for younger versus older adults. Proportions are shown for the magnitude of step sizes between consecutive trials (left) alongside the number of consecutive trials spent in given state (right). (D) Pupil diameter (Pupil diam.) was measured during an alternating grayscale (Gr-sc) light stimulus, from which ocular limits of dilation and constriction were estimated for each participant. Consistent with reported trends, the ocular limits of the pupil in response to light was significantly smaller in older adults than younger (two-sample t-test, P<0.001, N=102/72; grayscale). (E) PPAS, as measured relative to light responses, were broader and lower in older adults (two-sided Wilcoxon rank sum test, P=0.005, N=102/72). (F) Dependence of behavioral performance on PPAS during 0 dB SNR trials, as measured relative to the light response bounds. Error bars denote standard error.

Figure 3.

Pupil-indexed listening effort is greater in older adults but does not account for trial-to-trial variations. (A) Stimulus-evoked pupil responses were larger in more difficult listening conditions and in older adults (two-way ANOVA, N=102/72, main effect of SNR [F= 207.1, P<0.001], main effect of age [F=10.8, P=0.001], SNR × age [F=1.0, P=0.32]). (B) In the same way that we partitioned pre-stimulus pupil diameters into quintiles, stimulus-evoked pupil responses were assigned to one of five equally sized bins per person. (C, D) No relationship was observed between accuracy or response times (RT) with stimulus-evoked pupil responses.

Figure 4.

Deficits at low arousal levels correlate with self-reported listening experience. (A) Average digits-in-noise performance at 0 dB SNR and Speech, Spatial and Qualities of Hearing Scale (SSQ) scores (Pearson’s r=0.18, P=0.017, N=179). Blue circles denote adults younger than 50 years and orange squares denote adults 50 years or older. (B) A schematic of the regression model for SSQ scores. (C) T-statistics for each coefficient in the model predicting SSQ scores (pure-tone average [PTA], mean accuracy [Avg. acc.], mean change in accuracy between the 3^rd and 1^st pre-stimulus pupil quintiles [Δ lower], mean change in accuracy between the 3^rd and 5^th pre-stimulus pupil quintiles [Δ upper], mean evoked pupil response [Ev. pupil]). (D) Deficits at lower levels of pupil-linked arousal correlated with SSQ scores, shown after controlling for other model co-variates (covs.; Pearson’s r=0.27, P=0.002, N=170).