A new tensor factorization approach for convolutive blind source separation in time domain

In this paper a new tensor factorization based method is addressed to separate the speech signals from their convolutive mixtures. PARAFAC and majorization concepts have been used to estimate the model parameters which best fit the convolutive model. Having semi-diagonal covariance matrices for different source segments and also quasi static mixing channels are the requirements for our method. We evaluated the method using synthetically mixed real signals. The results show high ability of our method for separating the speech signals.