A Robust Joint GSS And Source Localization Algorithm For Robot Audition In Strong Reverberant Environment

The research on robot audition aims to study advanced techniques to help robots acquire acoustic information from the ambient environments. In many real situations, the environment may be adverse like being highly reverberant, and the microphone signal received by the robots may consist of a superposition of several sounds. Geometric source separation (GSS) algorithm uses prior geometric information to separate simultaneously present sound sources and thus is suitable for robot audition applications. However, the performance of GSS deteriorates as the reverberation time increases even if highly accurate sources locations are available, let alone classic source localization (SL) methods themselves can hardly perform well in reverberant environments. In this paper, a joint GSS and SL algorithm is proposed for robot audition applications in strong reverberant environments. This new method estimates the parameters of blind dereverberation (BD), SL and GSS alternatively, which is similar to the conditional separation and dereverberation (CSD) method to release the one-source assumption of many BD algorithms. Furthermore, the proposed method can also be used as a robust SL algorithm alone if necessary. Experimental results verified the robust performance achieved by the proposed algorithm.