Comparing approaches for predicting behavioural speech-in-noise performance using cortical responses to unattended stimuli

Abstract

The cortical tracking of the acoustic envelope is a phenomenon where the brain's electrical activity, as recorded by electroencephalography (EEG) signals, fluctuates in accordance with changes in stimulus intensity (the acoustic envelope of the stimulus). Understanding speech in a noisy background is a key challenge for people with hearing impairments. Speech stimuli are therefore more ecologically valid than clicks, tone pips, or speech tokens (e.g., syllables) for assessing hearing. However, it remains unclear whether EEG responses to speech provide an advantage in predicting speech intelligibility. This study aimed to assess the ability of cortical responses to speech and speech-related sounds to predict behavioural speech-in-noise performance in listeners with normal hearing when they are not attending to the stimuli.

Twenty native English-speaking adults with normal hearing (aged 18 to 40 years) participated in a speech reception task, listening to English Matrix sentences presented at signal-to-noise ratios (SNRs) of -15, -10, -5, 0, and ∞ (no background noise) dB, and then identifying the words they heard in the sentences. In the EEG experiment, the participants then listened to continuous speech, broadband noise modulated by the envelope of speech, and repeating short /da/ stimuli presented at the same SNR levels as in the Matrix test. For the latter, Auditory Late Response (ALR) was estimated from the EEG, and for the former, the strength of the envelope-tracking responses was calculated.

Cortical responses to all stimuli showed monotonic relationships with the signal-to-noise ratio at the group level and in most individuals, although there was considerable variability. EEG analysis in the delta band showed no significant difference in the number of participants with predicted speech reception thresholds (SRTs) within an error margin of 7 dB—the level at which SRT prediction is considered applicable—regardless of the type of cortical response used. In the theta band, however, SRT predictions based on cortical responses to continuous speech performed worse, showing a significantly lower number of predictions within an error margin of 7 dB compared to those based on cortical responses to modulated noise and the repeating /da/ sound. The proportion of individual SRT predictions with an error margin within 7 dB was, at best, 30 %.

For people with normal hearing, cortical responses to continuous speech and modulated noise predicted speech-in-noise performance at the group level but not at the individual level, due to variability in cortical tracking of the acoustic envelope. Predicting the SRT on an individual level remains a major and clinically important challenge.