Exploring Factors Important for Clinical Application of Cortical Responses to Continuous Speech

Abstract

There is considerable interest in neural responses to continuous speech. Techniques for analysing these responses typically involve tracking EEG change due to stimulus features, such as the amplitude envelope. However, the clinical utility of these measurements, especially for challenging to test subjects such as infants with hearing aids, remains under-explored. This thesis aimed to investigate the clinical feasibility of neural tracking as an objective test for aided speech detection in infants. This aim was tackled through four studies designed to test factors essential for future application in infant testing in clinical environments. These factors included the feasibility of detecting responses in single-channel EEG recordings, detection time, effects of stimulus intelligibility, and attention. The two approaches used to analyse EEG signals were the temporal response function (TRF) and cross-correlation. The first study assessed the effectiveness of single-channel EEG testing, achieving a 100% detection rate using cross-correlation within a detection time appropriate for clinical application. The second study focused on speech intelligibility effects during passive listening in recordings of cortical responses via single-channel EEG. The responses to speech-modulated noise demonstrated greater robustness regarding detectability and detection times than natural speech, indicating the potential utility of non-language-specific stimuli. Nevertheless, detection rates fell below 100%, potentially due to passive listening or shorter recording durations compared to the first study. The third study evaluated the envelope distortion induced by hearing aids using various stimuli. It found that the envelope distortion from the International Speech Test Signal (ISTS) was similar to that of natural speech, in contrast to speech-modulated noise, which exhibited significantly lower envelope distortion. The fourth study investigated the impact of different distortion levels on response detection using ISTS recordings from the third study. Higher levels of envelope distortion significantly lowered detectability and increased detection times, though using the envelope measured at the hearing aid output for detection analysis significantly improved these metrics. Additionally, no impact of attention on response detectability was observed. In conclusion, single-channel EEG analysis showed variable detectability across different stimuli and conditions, suggesting that signal processing methods and recording times may still need to be optimised. The ISTS stimulus produced results comparable to natural speech, supporting its potential clinical use as a non-language-specific option. However, detectability was compromised in aided condition with high levels of envelope distortion. Using the speech stimulus from the output of a hearing aid (as opposed to the input signal to the aid) shows potential for improving response detectability. Additionally, the study demonstrated that neural tracking can be recorded under passive listening conditions, which could be important when testing infants.