Tensor Factorization with Application to Convolutive Blind Source Separation of Speech

Decomposition of mixed information into its constituent components has been very useful in many applications such as acoustics, communications, and biomedicine. Eigenvalue decomposition (EVD), singular value decomposition (SVD), and independent component analysis (ICA) based on various criteria such as uncorrelatedness, independency, minimizing mutual information, and differences in distributions, have been widely used for this purpose before. In the case of convolutive mixtures however, further processing to handle multiple time lags of the signals has to be undertaken. Although many researchers have worked on convolutive blind source separation, as comprehensively reported by Pederson et al. in (Pederson et al., 2007) no robust solution, as given for linear instantaneous cases, has been reported. Moreover, generally, the uncorrelatedabStract