Comparing approaches for predicting behavioural speech-in-noise performance using cortical responses to unattended
Suwijak Deoisres1, Ghadah S Aljarboa1, Steven L Bell1
1Institute of Sound and Vibration Research, University of Southampton, Southampton, United Kingdom.
Hearing Research
|January 17, 2025
Summary
Cortical tracking of acoustic envelope using electroencephalography (EEG) shows promise for predicting speech intelligibility. However, individual-level prediction of speech reception thresholds (SRTs) remains challenging due to variability in brain responses.
Area of Science:
- Neuroscience
- Auditory Neuroscience
- Signal Processing
Background:
- Cortical tracking of the acoustic envelope, measured via electroencephalography (EEG), reflects brain responses to auditory stimuli.
- Understanding speech in noise is crucial for hearing impairment, with speech stimuli being more ecologically valid than artificial sounds for hearing assessment.
- The predictive power of EEG responses to speech for speech intelligibility remains unclear.
Purpose of the Study:
- To evaluate the efficacy of cortical responses to speech and speech-related sounds in predicting behavioral speech-in-noise performance in normal-hearing listeners.
- To assess if non-attended stimuli can yield predictive EEG responses for speech intelligibility.
- To determine if speech stimuli offer an advantage over modulated noise or /da/ stimuli in predicting speech reception thresholds (SRTs).
Main Methods:
- Twenty normal-hearing adults underwent a speech reception task using Matrix sentences at various signal-to-noise ratios (SNRs).
- EEG was recorded while participants listened to continuous speech, speech-envelope-modulated noise, and repeating /da/ stimuli at identical SNRs.
- Auditory Late Response (ALR) and envelope-tracking response strength were calculated from EEG data to predict SRTs.
Main Results:
- Cortical responses to all tested stimuli exhibited monotonic relationships with SNR at group and individual levels, despite considerable variability.
- In the delta band, no significant difference was found in the number of participants with predicted SRTs within a 7 dB margin across stimulus types.
- In the theta band, predictions based on continuous speech performed worse than those based on modulated noise or /da/ stimuli, with individual prediction accuracy at best 30%.
Conclusions:
- Cortical responses to continuous speech and modulated noise can predict speech-in-noise performance at a group level for normal-hearing individuals.
- Predicting SRTs at an individual level remains a significant clinical challenge due to variability in the cortical tracking of the acoustic envelope.
- Further research is needed to improve the reliability and accuracy of individual-level SRT prediction using EEG-based measures.
