Speech enhancement by generalized wavelet shrinkage

Abstract

The wavelet shrinkage denoising can effectively reduce the noise of non-stationary signal but preserve the local regularity. The key questions of wavelet shrinkage are how to choose shrinkage function and threshold value. A speech enhancement algorithm based on wavelet shrinkage is proposed. The generalized wavelet shrinkage functions are built and the Stein Unbiased risk estimate threshold value is derived. Noisy speech signals are used for the performance evaluation of the denoising algorithm. The optimal denoising scheme is achieved. The results indicate that the speech enhancement algorithm using the wavelet transform is promising. (4 pages)