Subcortical electrophysiological measures of running speech

Electroencephalography (EEG)-based objective measurements of neural responses to speech along the auditory pathway are important for research on speech processing as well as for clinical applications. Processing of sound in the auditory brainstem is traditionally assessed using many repetitions of short, simple stimuli that can not fully capture the complexity of speech. This thesis assesses a novel way of measuring midbrain processing to running, naturalistic speech by estimating temporal response functions (TRFs) using stimulus-response regression models. First, we bridge the gap between divergent published approaches by demonstrating that the interplay between model regularization and stimulus features causes morphological differences in response functions that must be dissociated from neural effects. In particular, we show that it is difficult to dissociate distinct pitch-related processing from responses to amplitude variations in the broadband speech signal when analyzing brainstem
responses to running naturalistic speech. We confirm that regression-modelled brainstem responses to broadband speech signal variations (speech-ABR) show a distinct positive peak at a latency of around 5-9 ms which approximates wave V of the conventional click-evoked auditory brainstem response (click-ABR) in latency and amplitude. Second, we address the test-retest reliability of the speech-ABR in the most numerous target group for hearing interventions, older hard of hearing (O-HOH) adults. Measuring across multiple recording sessions on different days, we show that the speech-ABR response peak is reliable in amplitude and latency across sessions when identical speech material is presented. Simultaneously assessed cortical TRFs do not show reliability across different days. We further confirm test-retest reliability of traditional subcortical and cortical measures, specifically, click-ABRs and cortical auditory evoked potentials (AEPs). Third, we apply the speech-ABR to quantify neural processing differences between younger normal-hearing (Y-NH) and O-HOH when compensating for effects of audibility with audiogram-based amplification or not. Listening to amplified speech, O-HOH showed delayed midbrain responses of reduced amplitudes. Amplitude differences of cortical TRF peaks corresponding to N1 and P2 of the AEPwere larger in O-HOH with and without providing amplification. For unamplified speech, midbrain wave V responses were practically gone in O-HOH, while the cortical activitymatched activity to amplified stimuli, suggesting the presence of a neural gain mechanism and demonstrating that enhanced neural tracking with age-related hearing loss is not driven by amplification. We further found that midbrain and cortex responses were positively correlated in amplitude and model prediction accuracy when O-HOH heard amplified speech, but not in Y-NH listeners, consistent with the notion that information redundancy may arise from peripheral deterioration. Finally, we discuss overarching questions, challenges, and future application perspectives in hearing research, diagnosis, and rehabilitation.