A robust correlation method for solving permutation problem in frequency domain blind source separation of speech signals

This paper addresses the well-known permutation problem, inconsistent permutation in the Discrete Fourier Transform bins after independent component analysis, in the frequency domain blind source separation. The method which utilizes the correlation between the adjacent bins in speech signals is popular among the techniques for solving permutation problem. However, the reliability of the method is very low. This paper presents a robust correlation method for solving the permutation problem, which utilizes the correlation between the signals in each DFT bin, one of which is partially separated by a time domain BSS method and the other is obtained by a frequency domain BSS method. The proposed method showed almost the same separation performance as that of the method which uses the correlation between adjacent bins and is highly reliable at the same time.