Speech enhancement based on combination of wiener filter and subspace filter

Abstract

This paper propose a novel multi-channel speech enhancement method by combining the wiener filtering and subspace filtering with a convex combinational coefficient. Because of using both the advantage in noise reduction of the subspace speech enhancement technology and the stable characteristic of the Wiener filtering technology, the proposed multi-channel speech enhancement method has a better performance in robustly removing colored noise from noisy speech signals. Simulation examples confirm that under different colored noise, the proposed multi-channel speech enhancement method can obtain better speech recovery results than the traditional subspace multi-channel speech enhancement method and the multi-channel Wiener filter speech enhancement method.