Speech enhancement using the combination of adaptive wavelet threshold and spectral subtraction based on wavelet packet decomposition

Abstract

A speech enhancement approach using adaptive wavelet threshold and spectral subtraction in the wavelet domain is proposed in this paper. First, in order to maintain the linear phase and prevent the spectral aliasing of reconstructed signals, the noisy signals are decomposed by un-decimated bi-orthogonal wavelet packet. Second, spectral subtraction is used for reducing noise from the two low-frequency sub-bands, which avoids the excessive distortion of low-frequency components caused by wavelet de-noising. Next, the improved wavelet shrinkage algorithm is adopted in the other high-frequency sub-bands. This threshold is updated by adaptively tracking the level of noise. Finally, these modified wavelet coefficients are reconstructed to obtain the enhanced speech signals by the inverse wavelet packet transform. The experimental results show that the proposed approach outperforms spectral subtraction method and wavelet de-noising method.