Modeling Sound Localization within the Framework of Bayesian Inference
* Presenting author
Abstract:
In this work, we propose an auditory model based on Bayesian inference to reproduce the individual human ability to localize a sound source in the acoustic free field. The model combines physiologically motivated front-ends with a probabilistic decision stage in order to estimate both the lateral- and polar-angle components of the incoming sound direction. In our systematical evaluation, the model was able to reproduce the summary statistics from five listeners when localizing broad-band sound sources. In particular, the results indicate that the model required to account for both the acoustic interaction of the sound source with the subject anatomy and for non-acoustic factors as neural uncertainties and sensorimotor mapping. On the other hand, we found little agreement between simulations and experimental data when considering distortions in the stimuli’s spectrum. This mismatch is probably related on how the auditory front-ends combined various spatial cues. We will discuss these results and further extensions of the framework to match actual performances in diverse acoustic scenarios.