Broadband Deterministic Blind Beamforming for Speech Separation Using Microphone Arrays via Rotational Invariance Techniques

In this paper a novel approach for speech separation using microphone arrays is presented. It belongs to the class of deterministic blind beamforming (DBBF) approaches via rotational invariance techniques. Since speech should be considered as broadband sources in microphone array related applications, the conventional narrowband DBBF is extended to broadband cases in this paper. Utilizing the nonstationarity of speech, DBBF is implemented at the frequency bins of interest. In order to cope with the problem of channel swap, a rearranging operation is performed, based on the correlation coefficients between the frequency domain separated sequences of adjacent bins. Moreover, the scale ambiguity, another drawback of the general blind source separation approaches, is also eliminated in this paper. Numerical examples confirm the effectiveness of the proposed approach.