A blind extraction of temporally correlated but statistically dependent acoustic signals

We propose a batch learning algorithm for sequential blind extraction of arbitrary distributed but generally not i.i.d. (independent identically distributed) temporally correlated sources, possibly dependent speech signals from from a linear mixture. The proposed algorithm is computationally very simple and efficient, it is based only on second order statistics and in contrast to most known algorithms developed for sequential blind extraction and independent component analysis, do not assume statistical independence of source signals nor non-zero kurtosis for the sources, thus statistical dependent signals including sources with extremely low or even zero kurtosis (colored Gaussian with different spectra) can be also successfully extracted. Extensive computer simulations confirm the validity and high performance of the proposed algorithm