BLIND SOURCE SEPARATION OF AUDIO SIGNALS USING WVD-KR ALGORITHM

Abstract

Under-determined blind source separation aims to separate N non-stationary sources from M (M<N) mixtures.Paper presents a time-frequency approach (TF) to under-determined blind source separation of N non-stationary sources from M mixtures(M<N). It is based on Wigner-Ville distribution and Khatri-Rao product. Improved method involves a two step approach which involves the estimation of the mixing matrix where negative values of auto WVD of the sources are fully considered and secondly auto-term TF points are extracted.After extracting the auto-term TF points source WVD values at every TF point are computed using a new algorithm based on Khatri-Rao product. Thus sources are separated with the proposed approach no matter how many active sources there are as long as N≤ 2M-1.Simulation results are presented to show the superiority of the proposed algorithm by comparing it with the existing algorithms.