Intelligent Sound Source Localization Based on Head-Related Transfer Functions

Abstract

While vision is a directed modality, i.e. constrained to a field of view (FOV), audition is not directed, and humans can localize sound outside their field. Similar to humans, robotic sound localization is of high importance, as it helps the robot to increase the degree of perception of its environment, especially in cases where vision is obstructed by barriers; this might be the case in many rescue scenarios either where the subject is hidden by collapsed structure or simply due to darkness. In this paper, we present a dual monaural-binaural sound localizer which is able to robustly localize sounds in 3D space. The algorithm isolates and detects the spectral shaping that the incoming sound signal experiences on its way to the microphones inserted inside the artificial ear canals of the robot. The proposed adaptive system was tested in simulations and further in a highly reverberant environment. Though deploying fewer microphones than state-of-the-art techniques, our proposed method achieved higher localization accuracy.