Detection by SNR maximization: Application to the blind source separation problem

Abstract

In this paper, we propose a method for the detection of stochastic signals, embedded in additive noise applied to the blind source separation problem in the particular case of delayed speech sources. The method proposed leads to a linear filter we call "constrained stochastic matched filter" (CSMF), which is optimal in the sense, that it maximizes the output signal-to-noise ratio (SNR) in a subspace whose dimension is fixed a priori. We show that the second-order statistics of sources can be unknown. © Springer-Verlag 2004.