Overdetermined Blind Separation of Acoustic Signals Based on MISO-Constrained Frequency-Domain ICA

ICA2004: 18th International Congress on Acoustics, April 4-9, 2004, Kyoto, Japan.We propose a new overdetermined blind source separation (SSS) using frequency-domain independent component analysis (FDICA) based on multiple-input single-output (MISO) constraint. To achieve a superior separation performance under reverberant environments, we set the number of microphones to be larger than that of sources. This leads to alternative problems in which the sound qualities of the separated signals are deteriorated. In order to solve these problems, we propose the overdetermined SSS based on MISO-constrained FDICA. The experimental results reveal that the sound qualities of the separated signals in the proposed method outperform that in the conventional FDICA especially in the large-array case