Blind signal separation for convolved nonstationary signals

Abstract

Blind signal separation is a signal processing technique that can extract the original signals from their mixtures by using only the observed signals when the mixed signals can be observed by multiple sensors. In this paper, we propose a technique for blind signal separation of separating original signals from the observed signals which mix the nonstationary signals, such as speech or music. The proposed method implements blind signal separation by minimizing a nonnegative function that achieves the minimum (zero) when the second-order cross-correlation value of the observed signals becomes zero. The effectiveness of the proposed method is verified by computer simulations and tests that used a mixed speech signal observed in an ordinary room. © 2000 Scripta Technica, Electron Comm Jpn Pt 3, 84(2): 1–11, 2001