A COMPRESSIVE SENSING APPROACH TO SPEECH SEGREGATION

Abstract

The problem of underdetermined blind source separation is usually addressed under the framework of sparse signal representation. This paper represents Compressive Sensing technique used for speech segregation that contains two stages. In the first stage we exploit a modified K-means method to estimate the unknown mixing matrix. The second stage is to separate the sources from the mixed signals using the estimated mixing matrix. In the second stage a two-layer sparsity model is used which assumes that the low frequency components of speech signals are sparse on K-SVD dictionary and the high frequency components are sparse on discrete cosine transformation (DCT) dictionary. In this way, we reconstruct the signals using L1-magic and GPSR algorithm.