Blind decomposition utilizing aggregated microphones

This paper describes a blind decomposition algorithm utilizing two directional microphones located at the same place, namely aggregated microphones. Recently, some authors proposed sound separation algorithm utilizing sparseness of the sounds. In these methods, we can extract each signal using the time-frequency binary masks. However, these method can not separate the sounds if the sounds overlap in frequency domain due to the sparseness assumption. In this paper, we treat the problem of separating not only the sounds which hardly overlap but also the sounds which overlap in most of frequency domain. It should be noted that we need to know neither the directivities of two microphones nor the overlapping frequency components in advance. In this paper, we describe the necessary conditions and the algorithm of the proposed method. Some experimental results show that the proposed method can separate not only the sounds which hardly overlap but also the sounds which overlap in most of frequency domain.