Information Theoretic Approaches to Source Separation

The problem of extracting sources from a mix of these has been extensively addressed by a lot of research. In this thesis the problem is being considered in the sonic domain using algorithms that employ information theoretic tech-niques. Basic source separation algorithms that deal with instantaneous mix-tures are introduced and are enhanced to deal with the real-world problem of convolved mixtures. The proposed approach in this thesis is an adaptive algorithm that operates in the frequency domain in order to improve efficiency and convergence behavior. In addition to developing a new algorithm there is also an effort to point important similarities between human perception and information theoretic computing approaches. Source separation algorithms developed by the audi-tory perception community are shown to have strong connections with the basic principles of the strictly engineering algorithms that are introduced, an