The role of canonical neural computations in sound localization
PhD thesis, Imperial College London (2019)
Localizing sounds is an important ability for many species. However, reverberative sounds present a signicant challenge to the auditory system as later arriving reverberations may carry confounding localization cues. The 'precedence eect' refers to a set of perceptual behaviours related to this situation. Studies investigating the precedence eect observed that the auditory system tends to focus the core of the localization process on the computation of localization cues carried by the rst arriving sound. Doing so relieves the auditory system from dealing with contradictory localization cues in later arriving sounds. A recent study by Dietz et al. (2013) conrmed that human listeners use this approach to deal with dynamic localization cues. In order to provide an explanation for this nding, we rst tested several auditory models on the specic task described in Dietz et al. (2013) in order to shortlist possible mechanisms capable of accounting for the early extraction of temporal binaural cues. We found that the best candidates to account for this data are single cell mechanisms, such as adaptation and onset ring, as well as inhibitory population mechanisms. To further understand how each mechanism contributes to the suppression of lagging sounds, we designed more general models capable of demonstrating the principal features of each mechanism. We tested these models thoroughly and found that all mechanisms were able to reproduce the results over a wide range of parameters. This nding suggests that mechanisms responsible for the precedence eect may not be specialized to perform this specic task but instead may be the results of more commonly found neural circuits in the brain. Finally, to facilitate comparing the performance of auditory models on psychoacoustical data, we also designed and implemented an auditory modelling framework capable of addressing many challenges existing in the eld of auditory modelling.
A Python simulator for spiking neural networks.
Python/Matlab package for comparing binaural auditory models.