複数マイクロホンアレイを用いたnmfによる空間音源分離法の提案と評価

Abstract

Sound source separation extracts only sound sources of interest from a mixture of sound sources and it is used as pre-processing for automatic speech recognition. For example, it reduces ambient noise, and automatic speech recognition and speaker identification are expected to improve. A commonly used sound source separation method is called beamforming using a microphone array consisting of multiple microphones. Although beamforming can separate sound sources based on the direction obtained from inter-microphone time and level differences, it has a limitation that it cannot separate sound sources in the same direction. In this paper, we propose a location-specific source separation method using multiple microphone arrays to solve this problem. In the proposed method, first, each microphone array separates a target sound source, and each separated sound includes other noise sources in the same direction of the target sound source mentioned above. Since the target sound source is included in all separated sounds, the proposed method extracts signals commonly included in the separated sounds to remove such noise sources. Preliminary results through numerical simulation showed the proposed method with non-negative matrix factorization as sound source separation worked properly.