Speech enhancement based on the combination of spectral subtraction and wavelet thresholding

The proposed speech enhancement system combining the spectral subtraction and wavelet thresholding techniques together is used to reduce the additive wideband noise. The objective of this paper is to research the adaptability and efficiency of the wavelet types and levels to the noises. Large amount of experiments were implemented in different schemes according to the types of the wavelet function, thresholding criterions and the modifying manners for the noise wavelet coefficients. Simulation results in terms of both objective and subjective evaluations show that the 5-layer db3 wavelet function using heursure criterion, incorporating with the first-layer coefficient been modified, is more efficient than other schemes of the wavelet thresholding denoising method designed in the experimental procedure.