Convex Hull Convolutive Non-negative Matrix Factorization Based Speech Enhancement For Multimedia Communication

In this paper, an effective speech enhancement method is proposed for the next generation multimedia communication system. The priori knowledge of the enhancement stage is obtained by the modified Convex Hull Convolutive NMF with less information loss. To deal with the difficulty of its optimal gain estimation, an iterative algorithm is then introduced to update the coefficient matrix. The experimental results under different types of noise environment show that the proposed algorithm can reduce the signal distortions dramatically, and provide better enhancement performance than the benchmark algorithms simultaneously, especially under adverse conditions.