Localization, Tracking, and Separation of Sound Sources for Cognitive Robots

Abstract

An auditory system plays an important role for the perception system of cognitive robots. This thesis investigates auditory system modules for the localization, tracking, and separation of sound sources. The presented localization module operates in the time-frequency domain and exploits signal sparseness to estimate the positions of multiple sources. The tracking module uses particle filters with bimodal observation probability densities to post-process the localization results. The separation module is based on binary masking and shares its computations with the localization module to keep its complexity low. Real-world experiments reveal that the presented algorithms perform better than state-of-the-art techniques, while simultaneously operating inside the requirements and constraints of robotic systems.