Two-step Technique Based Laplacian Factor Estimation for Speech Enhancement

The estimation of Laplacian factor is a crucial part of speech enhancement algorithms using Laplacian model priori. Classical methods for the estimation of this parameter suffer from the residual noise or time delay bias. In this paper, a novel algorithm called two-step technique for the estimation of Laplacian factor is proposed in discrete cosine transform domain. Based on the estimation theory of minimum mean square error (MMSE), the proposed algorithm computes directly the amplitude of the clean speech component to refine the estimated Laplacian factor in the original method, and thus the accurate result can be attained. Simulation results under different kinds of noisy conditions and levels demonstrate that the proposed algorithm possesses improved performance than that of the original method.