Improved method for solving permutation problem of frequency domain blind source separation

Abstract

The permutation indetermination is a very important problem in the frequency domain blind source separation. This paper proposes a novel algorithm based on amplitude correlation between two adjacent frequency bins. The amplitude correlation between adjacent frequency bins of the same speech signal is better than that of different speech signals. We construct amplitude correlation matrixes between two adjacent frequency bins according to the amplitude correlation, and then map the independent components (ICs) to the corresponding sources. The proposed method is on-line. It is feasible to implement ICA algorithm and proposed method simultaneously. Simulation results showed the encouraging performance.