Sparse filter models for solving permutation indeterminacy in convolutive blind source separation

Abstract : Frequency-domain methods for estimating mixing filters in convolutive blind source separation (BSS) suffer from permutation and scaling indeterminacies in sub-bands. Solving these indeterminacies are critical to such BSS systems. In this paper, we propose to use sparse filter models to tackle the permutation problem. It will be shown that the ℓ1-norm of the filter matrix increases with permutations and with this motivation, an algorithm is then presented which aims to solve the permutations in the absence of any scaling. Experimental evidence to show the behaviour of ℓ1-norm of the filter matrix to sub-band permutations is presented. Then, the performance of our proposed algorithm is presented, both in noiseless and noisy cases.