Multiple Sound Sources Localization by using Statistical Source Component Equalization

Abstract

Multiple sound sources localization is a hot topic in audio signal processing as it can provide effective information for parameter coding and reconstruction of sound scenes. In this paper, a multiple sound sources localization method is proposed by using statistical source component equalization. Based on single source zone (SSZ) detection, the proposed method aims to settle the localization accuracy degradation problem caused by the missed detection of statistically weak source (SWS) which is inevitable in the sound scene where more than five sound sources occur simultaneously. Since SWSs only have little DOA estimations compared with other sound sources called statistically dominant source (SDS), they are difficult to be found in the histogram of DOA estimations. A statistical source component equalization algorithm is designed to remove the components of SDSs and reserve the components of SWSs at the same time, which can make the SWSs obvious enough to be found through post-processing. The objective evaluation reveals that the proposed method can always obtain a comparable or better localization results than traditional SSZ-based method