Blind Source Separation Based on Multistage ICA Using Frequency-Domain ICA and Time-Domain ICA

ICFS2002: The International Conference, on Fundamentals of Electronics Communications and Computer Science of IEICE, March 27-28, 2002, Tokyo, Japan.We propose a new algorithm for blind source separation (BSS), in which frequency-domain independent component analysis (FDICA) and time-domain ICA (TDICA) are combined to achieve a superior source-separation performance under reverberant conditions. In the proposed method, the separated signals of FDICA are regarded as the input signals for TDICA, and we can remove the residual cross-talk components of FDICA by using TDICA. This can improve both learning stability in ICA and source-separation performance. The experimental results under the reverberant condition reveal that the separation performance of the proposed method is superior to those of the conventional ICA-based BSS methods