Characterising the Human Auditory System using a Linear Least-Squares System Identification Approach

Abstract

Disabling hearing loss affects many millions of people around the world.Early identification and suitable interventions, e.g., the provision of hearing aids, cochlear implants, etc., can help but are limited by the methods currently used to assess hearing function.Hearing function can be assessed in one of two ways: using either subjective or objective methods.Subjective methods rely on the behavioural response to sound, while objective methods rely on the physiological response to sound.While subjective methods such as pure tone audiometry (PTA) have played an important role in hearing assessment for decades, they are generally limited to simple detection tasks, e.g. detecting pure tones inquiet, and provide little indication as to the source of any deficit.While some objective measures such as the auditory evoked potential (AEP)—a transient electrophysiological response to sound—can provide more detailed neurophysiological information, they are often restricted by the need to use simple discrete stimuli such as clicks or tone-bursts, which are arguably not that representative of everyday sounds.In recent years, there has been growing interest in the use of modelling approaches to study and assess the human auditory system.One approach that has proven particularly useful is temporal response function (TRF) estimation. With TRF estimation, the assumption is that the output neural data, consists of the convolution of some input stimulus feature with an unknown system response, plus noise.Given the known stimulus feature and the recorded neural response,the intermediary system response(TRF), can be derived using TRF estimation. One of the main advantages of this approach is that it provides similar responses to the AEP, while permitting a wider variety of stimuli to be used.Much of the work that has been done using this approach has been focused on the study and assessment of high-level(cognitive)auditory processing. The overall aim of this thesis is to develop and appraise new methodological approaches that facilitate the use of TRF estimation in the study and assessment of low-level (sensory)auditory processing. In Chapter 4, two approaches for indexing low-level processing along the auditory pathway are introduced over two experiments: Experiment 1 and 2. Experiment 1 is an initial exploratory attempt to derive responses along the auditory pathway to click trains,i.e., sequences of click stimuli—the classic stimuli of auditory research. Experiment 2 is a more thorough attempt to derive responses along the auditory pathway to amplitude modulated (AM) broadband noise (BBN), using a novel efficient TRF estimation approach. Considerations of stimulus type, stimulus representation, i.e., what stimulus feature to use and howto represent it in the analysis,and computational efficiency are discussed, and the neural underpinnings of the derived responses investigated through comparisons with their canonical counterparts, i.e., AEPs, elicited using chirp trains.In Chapter 5, a novel TRF estimation approach for objectively determining hearing thresholds, i.e., the lowest levels at which certain sounds can be heard in quiet, using multiplexed, i.e., multiple, mixed, AM tones (AMTs) is presented. Considerations of stimulus type, stimulus representation, and modelling approach are discussed, and the performance of this approach evaluated through comparisons with thresholds recovered using PTA, in both normal hearing and hearing loss populations.In Chapter 6, several novel stimulus representations are presented with the view of enhancing response derivation using TRF estimation. The importance and benefits of taking certain neurophysiological properties of the human auditory system into account when designing stimulus representations are discussed and then quantified through comparisons with other models derived using more standard stimulus representations.While the focus of this thesis is on low-level auditory processing, any high-level processing necessarily involves some degree of low-level processing as well. Therefore, it is hoped that the work presented in this thesis will help to further the study and assessment of both low-and high-level processing using TRF estimation. As such, it represents an opportunity to move the field closer towards more diagnostically useful objective measures of hearing function along the auditory pathway that can be implemented using a wider variety of stimuli.