A New Method of Solving Permutation Problem in Blind Source Separation for Convolutive Acoustic Signals in Frequency-domain

Abstract

This paper proposes a novel scheme to solve permutation ambiguity in frequency-domain for the separation of the convolutive mixing signals. We use sparseness of original acoustic signals to build a histogram of direction of arrival (DOA) for the original signals, and then use mask technique to get rough recovered signals with distortion but no order problem. An independent component analysis (ICA) is implemented to solve more accurate separation at each frequency bin. The permutation problem can easily be solved based on the rough recovered signals by mask of DOA histogram. Compared with the existing algorithms, the proposed algorithm has better performance than both ICA and time-frequency mask methods.