Auditory cortical responses as an objective predictor of speech-in-noise performance

Abstract

The cortical entrainment of speech envelope is a phenomenon where the electrical activity in the brain fluctuates with the change in stimulus intensity, i.e., the stimulus envelope. While clearly speech stimuli are closer to everyday conversation, it is not clear whether the cortical entrainment of speech envelope closely reflects speech intelligibility in the brain more than cortical responses to other types of sound. The overall goal of this thesis is to assess the applicability of cortical responses to speech and non-speech sound for use as a predictor of behavioural speech intelligibility in normal hearing people. Initial works were carried out to explore how to best measure and detect cortical responses to continuous speech.

In the first study cortical responses to continuous speech were stronger when additional pauses were inserted into the stimulus, and this consequently led to a greater the number of detected responses. This also demonstrate that the amount of pauses could introduce a comparison bias as they have different envelopes, it was therefore decided that the continuous speech and non-speech stimuli used in the third study would comprise identical envelope shapes. In the second study, the backward linear modelling (reconstructing speech envelope from cortical responses) was able to detect greater number of significant responses to continuous speech than the forward linear modelling (predicting cortical responses from speech envelope), however, the number of detected responses was lower than the cortical auditory evoked potentials (CAEP) using Hotelling’s T-squared. This suggests that evoked responses to continuous speech are harder to detect than more standard evoked responses. In the third study, behavioural speech recognition scores and objective measures showed significant correlation and it was possible to utilise cortical responses to continuous speech, modulated noise, and /da/ as objective measures to predict speech intelligibility in some subjects. However, the reliability of response measurement was poor, consequently the speech intelligibility prediction was inaccurate in many individuals.

With regards to the main goal of this thesis, all types of cortical responses showed expected monotonic relationship with the stimulus signal-to-noise ratio on a group level, cortical responses to continuous speech were worse in predicting the speech intelligibility in individuals than other types of responses. This gives empirical evidence to show that measurement of cortical responses to continuous speech through the linear modelling approach with scalp responses measurement are less reliable compared to cortical responses to less natural speech stimuli at an individual level. For normal hearing people, a measurement that simply confirms audibility (not necessarily speech intelligibility or comprehension) may be sufficient to predict speech-in-noise test performance.

More information

Identifiers

ORCID for Suwijak Deoisres: orcid.org/0000-0003-1384-3632

ORCID for David Simpson: orcid.org/0000-0001-9072-5088

Catalogue record

Export record