Auditory skills and brain morphology predict individual differences in adaptation to degraded speech

Abstract

Noise-vocoded speech is a spectrally highly degraded signal, but it preserves the temporal envelope of speech. Listeners vary considerably in their ability to adapt to this degraded speech signal. Here, we hypothesised that individual differences in adaptation to vocoded speech should be predictable by non-speech auditory, cognitive, and neuroanatomical factors. We tested 18 normal-hearing participants in a short-term vocoded speech-learning paradigm (listening to 100 4-band-vocoded sentences). Non-speech auditory skills were assessed using amplitude modulation (AM) rate discrimination, where modulation rates were centred on the speech-relevant rate of 4 Hz. Working memory capacities were evaluated (digit span and nonword repetition), and structural MRI scans were examined for anatomical predictors of vocoded speech learning using voxel-based morphometry. Listeners who learned faster to understand degraded speech also showed smaller thresholds in the AM discrimination task. This ability to adjust to degraded speech is furthermore reflected anatomically in increased grey matter volume in an area of the left thalamus (pulvinar) that is strongly connected to the auditory and prefrontal cortices. Thus, individual non-speech auditory skills and left thalamus grey matter volume can predict how quickly a listener adapts to degraded speech.

Introduction

Noise-vocoding has been used to simulate cochlear-implant (CI) transduced speech in normal-hearing listeners (Shannon, Zeng, Kamath, Wygonski, & Ekelid, 1995). Vocoding removes most of the spectral properties of the auditory signal, while its temporal envelope cues remain largely preserved. Normal-hearing listeners learn to understand this type of degraded speech surprisingly quickly on the first exposure (Davis et al., 2005, Eisner et al., 2010). However, there is considerable variability across listeners in their ability to perceptually adapt to this degraded auditory input. Although there are larger individual differences in speech perception performance in CI users than in normal-hearing subjects listening to a CI simulation, normal-hearing listeners still differ considerably in learning to understand vocoded speech (Shannon, Galvin, & Baskent, 2002). Currently it is unclear what drives the adaptation to this degraded speech signal. In cochlear-implanted children, demographic factors such as age at implantation, duration of deafness, residual hearing before implantation or the number of electrodes of the CI can only explain about half of the variance in outcome (Blamey et al., 2001, Sarant et al., 2001).

Phonological working memory is one cognitive factor that has been linked to the development of speech recognition abilities (Gathercole, Willis, Baddeley, & Emslie, 1994). According to the model proposed by Baddeley and colleagues, working memory relies on two processes: the “buffer”, which holds memory traces for a few moments, and the “phonological loop”, which can be imagined as a subvocal rehearsal process (Baddeley, 2012). Pisoni and colleagues have investigated the relationship between phonological working memory and individual speech recognition performance in CI users, and found that working memory scores, as measured by digit span, significantly correlated with spoken word recognition scores in paediatric CI users, even after statistical partialling out of demographic factors (Pisoni & Cleary, 2003).

In the present study, we wanted to examine whether we could replicate these results in normal-hearing adults listening to vocoded speech. Digit span requires the participant to hold online representations of perceived auditory information, taxing the “buffer” in Baddeley's working memory model. However, as digit span demands memorizing of highly familiar items that have a long-term semantic representation, it may not be a reliable test of phonological working memory (Jacquemot & Scott, 2006), especially if semantic and phonological representations have separable stores as some patient studies suggest (Romani & Martin, 1999). In order to test more specifically phonological working memory, we therefore additionally employed a nonword repetition test in the current study. Degraded speech recognition challenges phonological analysis and requires the maintenance of auditory input in the working memory for a short period of time. Thus, we expected that both tests of working memory, digit span and nonword repetition, should independently explain some of the variability associated with vocoded speech perception.

From an acoustic perspective, noise-vocoded speech has very poor spectral resolution. This forces listeners to rely more on temporal envelope cues for speech recognition. The utilisation of these cues might be easier for listeners with a higher sensitivity to envelope fluctuations in an auditory signal. We therefore hypothesised that the individual capability to adapt to vocoded speech is related to amplitude modulation (AM) rate discrimination thresholds. AM discrimination performance provides a measure of the auditory system's ability to encode temporal information in a waveform's envelope (Wakefield & Viemeister, 1990). The most prominent low-frequency amplitude modulations in the temporal envelope of speech are near the syllabic rate of approximately 4 Hz (Giraud et al., 2000, Houtgast and Steeneken, 1985). As these low AM frequencies are essential for speech perception, we decided to test listeners in their sensitivity to AM rates centred on 4 Hz. In the present study, we deliberately chose to assess sensitivity to AM rate rather than AM depth, which is traditionally used to determine temporal modulation transfer functions (Viemeister, 1979). We assessed thresholds to modulation rate instead, as it captures the temporal aspect of envelope encoding which is emphasised when listening to vocoded speech.

Hence we anticipated that, complementarily to phonological working memory scores, auditory processing skills that are not speech-specific (i.e., AM rate sensitivity) could account for part of the individual differences in perceptual learning of vocoded speech.

A small number of studies have investigated the relationship between brain activation patterns and perceptual adaptation to degraded speech. Notably, Adank and Devlin (2010) conducted a functional MRI study of adaptation to time-compressed speech. They observed adaptation-related activity bilaterally in the auditory cortex (posterior superior temporal sulcus) and in the left ventral premotor cortex. The authors concluded that perceptual learning of degraded speech involves mapping novel acoustic patterns onto articulatory motor plans (Adank & Devlin, 2010).

Using a short-term vocoded speech-learning paradigm, Eisner et al. (2010) found that individual performance improvement in vocoded speech perception was correlated with the functional MRI signal change in the inferior frontal gyrus (IFG). Additionally they observed a positive correlation between activity in the angular gyrus (AG) and individual learning curves over the course of 100 noise-vocoded sentences (but did not have continuous measures of performance throughout the scan period).

Golestani and colleagues explored the structural brain correlates of novel speech sound learning. They found that faster phonetic learners had increased white matter density in the parieto-occipital sulcus (Golestani, Paus, & Zatorre, 2002). Though, to our knowledge, structural correlates of vocoded speech learning have not been investigated so far. Here, we examine whether it is possible to predict perceptual learning of a vocoded speech from the brain structure of a listener.

The present study tested normal-hearing adults in a short-term (ca. 20 min long) vocoded speech-learning paradigm. A set of tests including digit span, nonword repetition and AM rate discrimination were administered. Additionally, we analysed structural MRI scans using voxel-based morphometry (VBM). The aim of the study was to elucidate whether and to what extent these cognitive, non-speech auditory, and neuroanatomic measures could predict how well a listener perceptually adapts to vocoded speech.