A method of ICA in time-frequency domain


We propose a method of ICA for separating convolutive mixtures of acoustic signals. The acoustic signals recorded in a real environment are not instantaneous but convolutive mixtures, because of the delay and the reflections. In order to separate these signals, it is effective to transform the signals into time-frequency domain. The difficult point in these approaches is the ambiguity of the permutation and amplitude which is unavoidable in original ICA problem. Since the basic ICA approaches cannot solve these ambiguity, we need another approach to solve them. We employed the envelopes of the signals to solve it, and have developed some algorithms. In this article, we show the outline of our orignal method, and some extensions of it. They are, the on-line version and auditory scene analysis problem.

Proceedings of International Workshop on Independent Component Analysis and Blind Signal Separation (ICA'99)