An improved combination of Directional BSS and a source localizer for robust source separation in rapidly time-varying acoustic scenarios

In rapidly time-varying acoustic scenarios, Blind Source Separation (BSS) often suffers from problems with slow convergence and poor separation performance. In this paper, we propose an improved scheme composed of Directional BSS and a source localizer for robustly separating and quickly tracking sources in a rapidly time-varying scene. Directional BSS is defined as a generic BSS algorithm combined with geometric constraints which are roughly equivalent to a set of delay&subtract beamformers. This scheme combines benefits from BSS and null beamforming and achieves high performance for separating sources of known locations. For estimating the Time-Difference-Of-Arrival (TDOA) of each source, another BSS system running in parallel to the main Directional BSS is exploited to serve as a source localizer. Experimental results illustrate the efficiency of the proposed concept.