Abstract
Over the decades, Bayesian statistical inference has become a staple technique for modelling human multisensory perception. Many studies have successfully shown how sensory and prior information can be combined to optimally interpret our environment. Because of the multiple sound localisation cues available in the binaural signal, sound localisation models based on Bayesian inference are a promising way of explaining behavioural human data. An interesting aspect is the consideration of dynamic localisation cues obtained through self-motion. Here we provide a review of the recent developments in modelling dynamic sound localisation with a particular focus on Bayesian inference. Further, we describe a theoretical Bayesian framework capable to model dynamic and active listening situations in humans in a static auditory environment. In order to demonstrate its potential in future implementations, we provide results from two examples of simplified versions of that framework.
Highlights
Sound localisation is a primary function of the human auditory system
; the posterior probability density function (PDF) p(w|y) of direction w of a source given acoustic information y depends on three factors: (1) The likelihood p(y|w), representing the PDF of acoustic information y being observed for a source at direction w; (2) The prior PDF p(w), representing assumptions on the result, derived from the past experience on the parameter to be estimated; and (3) The denominator p(y), representing the PDF of acoustic information y being observed andRassumed to be a normalisation constant inferred from pðwjyÞdw 1⁄4 1, so that the area under the posterior PDF integrates to 1
This article briefly reviews the recent literature on modelling active dynamic sound localisation, which complements the well-documented sound localisation based on static acoustic features with cues from self-motion or source motion
Summary
Sound localisation is a primary function of the human auditory system. Besides the well established evolutionary advantages [1], it is a crucial process for attention control and self-orientation. Monaural spectral cues, which result from the filtering properties of the outer ear, head and torso, carry additional information on the polar position of the source (Fig. 1). This spectral information aids in resolving the ambiguity in the binaural cues [5]. Multiple sound localisation cues available in the binaural signal, sound localisation models based on Bayesian inference seem to be a promising way of explaining behavioural human data. We describe a recursive theoretical framework for dynamic listening through Bayesian inference This framework aims at modelling dynamic listening situations which involve stationary sound sources in combination with head movements
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
