Auditory perception and localization are fundamental tasks for many species, allowing them to detect, identify, and spatially localize sound sources in their environment. While biological systems have evolved sophisticated neural mechanisms for auditory adaptation, artificial auditory systems still struggle to match their performance, particularly in dynamic and noisy environments. Our research focuses on whether sensor adaptation, driven by efferent feedback from the processing stage to the sensory stage, can improve localization performance. Inspired by human sound source localization based on interaural level differences (ILD) and efferent feedback, the proposed neuromorphic system architecture is composed of two bio-inspired acoustic sensors connected to a neural processing stage, represented by two neurons of the medial nucleus of the trapezoid body (MNTB) and two neurons of the lateral superior olive (LSO). The LSO neuron response was analyzed in the following ways: (i) using measured sensor responses at different ILD without efferent feedback and with a fixed local feedback for each sensor measurement; (ii) simulated with synthetically generated sounds with varying ILDs for four different feedback configurations from the LSO neuron to the acoustic sensors. Results from (i) showed how the feedback tuning can be used to overcome mismatches due to fabrication tolerances between different MEMS sensors, and (ii) showed the influence of different feedback configurations and simulation parameters on the LSO neuron response with respect to different ILDs.
Durstewitz et al. (Wed,) studied this question.