A second order multi output deconvolution (SOMOD) technique

In this paper, we present an efficient solution to the blind multi-channel deconvolution problem that consists of recovering independent source signals from their convolutive mixtures. In the case of instantaneous mixtures, a robust solution referred to as second order blind identification (SOBI) has been proposed previously. It is based on the joint diagonalization of spatio-temporal correlation matrices. Herein, we extend this technique to the convolutive mixture case. In contrast to existing deconvolution techniques, this new approach is able to deal with an overestimated source number. The proposed method has been successfully applied to the deconvolution of speech signals.