Blind separation of convolutive mixtures using joint block diagonalization

We present an efficient and new solution to the recovery of independent source signals from their convolutive mixtures without any a priori knowledge on the mixtures. In the case of instantaneous mixtures, a robust solution referred to as second order blind identification (SOBI) has been proposed previously. We extend this technique to the convolutive mixture case and we propose an algorithm that performs a joint block diagonalization of spatio-temporal correlation matrices instead of joint diagonalization as in SOBI. The proposed method has been successfully applied to the separation of speech signals.