A new method for underdetermined convolutive blind source separation in frequency domain

We proposed a new method for underdetermined convolutive blind sourece separation and permutation alignment. First, we combine the full-rank spatial covariance model with time-frequency masking to separate fourier transform coefficients of speech signals at each frequency bin, which can reduce computational complexity. After converting the variance parameters to binary masking sequences, we choose only the representative sequences for clustering. Finally, the artificial bee colony algorithm(ABC) is used to implement permutation alignment. The computer results show that the proposed method works very well compared with exising method.