Assessing and reducing listening effort of listening to speech in adverse conditions

Normally hearing listeners are skilful speech-perceivers, even in challenging listening environments. However, the neural processes required to compensate for missing or masked speech information can lead to tiredness and fatigue, even in everyday environments such as call-centres or shopping malls. This study utilises electroencephalography (EEG) to measure this cognitive compensation – termed listening effort (LE) – and its neurophysiological correlates, by comparing unprocessed speech to speech enhanced with AdaptDRC.

AdaptDRC is a near-end-listening enhancement (NELE) algorithm that alters speech signals for playback, dependent on environmental noise, and significantly improves speech intelligibility (Schepker et al., 2013). AdaptDRC reduces the subjectively rated effort for listening to speech in noise, even at 100% intelligibility (Rennies et al., in press) but we do not yet have corresponding neurophysiological objective data.

In this study, we recorded EEG while participants performed a LE task (N=30; normal hearing adults). Participants listened to unprocessed or AdaptDRC-processed OLSA sentences (Wagner et al., 1999) in cafeteria or speech-shaped noise at five signal-to-noise ratios (-10, -5, 0, +5, +10dB), then rated the effort required to understand the speech using a modified adaptive categorical LE scale (ACALES; Krueger et al., 2017). We also measured speech intelligibility: at intermittent trials, participants were prompted to repeat the sentence aloud, to ensure that participants were actively listening to the speech. Further, we measured participants’ hearing (pure tone audiometry) and cognitive abilities (working memory, selective attention, inhibition), to explore the relationship between individual differences in these abilities and subjective and objective LE.

Preliminary data analyses (N=12) indicate that, independent of noise type, speech intelligibility is at ceiling (98%) at SNRs of 0, +5 and +10 dB. At these SNRs, subjective LE decreases with increasing SNR (r=-.51) and AdaptDRC speech was rated lower than unprocessed speech. EEG analysis will identify neurophysiological markers of compensatory LE changes, and their relationships with subjective ratings. We focus on spectral power within the alpha frequency band (8-12 Hz), as work has shown a relationship between alpha spectral density and the suppression of task irrelevant information.

Thus, this experiment provides insight into the neurocognitive correlates of LE, the compensatory processes required for successful speech perception in sub-optimal conditions, and the benefits of speech enhancement technologies. The continued development and implementation of NELE technology in public and workplace environments will aid speech perception in suboptimal conditions and may also improve listener experience at low levels of noise.