Performance evaluation of blind source separation algorithms base on source signal statistics in convolutive mixtures

In this paper, we have investigated the performance of algorithms, based on second and higher-order statistics, which are developed for blind source separation (BSS) of acoustic signals. Acoustic signals are recorded simultaneously in a reverberant environment and can be described as sums of differently convolved signals. We have performed the performance of the blind algorithms in term of signal to interference ratio (SIR) and the choice of optimal separation filter length and the convergence of both algorithms are studied as well. The results can be used to work out the applicability as well as the parameters setup of such kind of blind algorithms.