Neural Correlates of Speech Intelligibility

Hearing loss is a growing public health concern, with many individuals struggling to understand speech in noisy conditions. The standard approach to evaluate speech understanding in noise is to measure the Speech Reception Threshold (SRT) behaviorally, which requires an active response. This PhD project aims to investigate the estimation of SRT using electroencephalography (EEG), allowing for a more automatic approach. Such an approach could enable SRT measurements to be incorporated into hearing aids, optimizing noise reduction, continuously monitoring users’ speech intelligibility, or even allowing at-home SRT assessments. <br/><br/>To estimate the SRT value from EEG, two methods were explored: one using the envelope of the stimuli reconstructed from the EEG response to continuous speech (SRTneuro) and the other using a neural component (SRTP300). For the SRTneuro estimation a linear decoder was used to reconstruct the envelope at different signal-to-noise ratios. A sigmoid function was then fitted to the reconstruction accuracies and the midpoint of the function was used as the SRTneuro. This method was used in younger (YNH) and older (ONH) normal hearing individuals. Within these populations, both scalp and in-ear electrodes were used for SRTneuro estimation. The results showed that the SRTneuro estimation quality for the ONH in general was similar to the YNH population, despite an overall increase in reconstruction accuracy observed in the ONH population. Furthermore it was seen that SRTneuro can be estimated in a subgroup of the tested populations using only EEG in the ear and that the estimation quality improved when combining in-ear electrodes with electrodes around the ear. To better understand the increase of reconstruction accuracies, the temporal response functions were compared, revealing an enhancement of amplitudes in the ONH population. <br/><br/>A potential problem with the SRTneuro method is that the reconstruction accuracy of the envelope is not a direct measure of speech intelligibility. Therefore the second method estimating SRTP300 was explored. Pilot-work and initial results of developing a paradigm to evaluate speech intelligibility based on a cognitive component is included in the PhD thesis. The results were inconclusive and future investigations should be conducted to utilize the paradigm for a SRTP300 determination.