Blind separation of noisy mixtures of non-stationary sources using spectral decorrelation

Abstract

In this paper, we propose a new approach for blind separation of noisy, over-determined, linear instantaneous mixtures of non- stationary sources. This approach is an extension of a new method based on spectral decorrelation that we have recently proposed. Contrary to classical second-order blind source separation (BSS) algorithms, our proposed approach only requires the non-stationary sources and the stationary noise signals to be instantaneously mutually uncorrelated. Thanks to this assumption, it works even if the noise signals are auto-correlated. The simulation results show the much better performance of our approach in comparison to some classical BSS algorithms. © Springer-Verlag Berlin Heidelberg 2009.