Multiple-sound-source localization scheme based on feedback-architecture source separation

Abstract

The problem of multiple-sound-source localization in a reverberant environment is investigated in this work, which is a critical technique in video conference applications and is still a challenging task in audio signal processing. In this study, sound signals are first estimated and separated from the mixtures received by a microphone array under a feedback architecture using the blind source separation algorithm. The parameters of time delay of arrival associated with microphones are calculated from the mixtures and estimated source signals by the time-domain cross correlation technique. Finally, the positions of sound sources are derived by the spatial position determination formulas. Experimental results demonstrate that the proposed scheme can effectively determine the locations of multiple sound sources in a near-field and reverberant environment. The direction prediction using the proposed scheme is very promising.