Research on decomposition level selection of wavelet shrinkage denoising

Threshold de-noising in wavelet domain is an efficient algorithm to reduce the white noise in digital signals. Even though much work has been done in this field, most of it was focused on the optimal choice of the threshold and usually selected a fixed decomposition level depend on experience. In this paper, we find that the appropriate decomposition level is anther key factor pertinent to de-noising performance. Then we propose a novel wavelet de-noising method for speech enhancement, which can adaptively select the optimal decomposition level for noisy speech under different SNRs. This new algorithm can adaptively choose higher decomposition level for low SNR to achieve more SNR gains and choose lower decomposition level for high SNR to prevent. over-thresholding. To evaluate the de-noising performance, we compare this method with the conventional one based on fixed decomposition level. The experimental results demonstrate that this proposed algorithm outperforms the classical wavelet-based de-noising method.