A New Approach for Blind Separation of Convolutive Mixtures

Abstract

An algorithm for blind source separation (BSS) of convolutive mixtures is discussed here. Separation of signals is performed in two stages. The first stage involves the application of an independent component analysis (ICA) algorithm and in the second stage shrinkage functions are applied to a set of wavelet coefficients. The ICA utilizes maximization of entropy to update the network weights and feedback is used within the network architecture. The ICA network alone can achieve acceptable levels of separation of artificially convolved sources. However, separation quality deteriorates for real-world convolutive mixtures. Hence, the separated signals can have cross-talk components. This work deals with the cross-talk problem by applying a novel postprocessing technique. The signals separated by the ICA network are passed through a post-processor, which has a set of shrinkage functions. The algorithm reduces the cross-talk components significantly as compared to using only the ICA algorithm.