UNDERDETERMINED BLIND SPEECH SEPARATION WITH DIRECTIVITY PATTERN BASED CONTINUOUS MASK AND ICA

We propose a method for separating speech signals when sources outnumber the sensors. In this paper we mainly concentrate on the case of three sources and two sensors. Some existing methods employ binary masks to extract the signals, and therefore, the ex-tracted signals contain loud musical noise. To overcome this prob-lem, we propose the utilization of a directivity pattern based contin-uous mask, which removes a single source from the observations, and independent component analysis (ICA) to separate the remain-ing mixtures. Experimental results show that our proposed method can separate signals with little distortion even in a real reverberant environment of TR=130 ms. 1