Blind Source Separation and Doa Estimation Using Small 3-d Microphone Array

We present a prototype system for Blind Source Separation (BSS) of many speech signals. Our system uses 8 microphones located at the vertexes of a 4cm×4cm×4cm cube and has the ability to separate signals distributed in three-dimensional space (Fig. 1). The mixed signals observed by the microphone array are processed by Independent Component Analysis (ICA)[1] in the frequency domain and separated into a given number of signals (up to 8). The system estimates the directions of arrival (DOA) of the source signals as a by-product of the separation process. In our previous work [2], we described the separation of 6 source signals of simulation data, i.e. signals made by convolving impulse responses. In contrast, this prototype system performs an on-the-spot BSS of live recorded signals. The system specifications are summarized in Table 1. The key technologies used in our system are as follows.