A versatile speech enhancement system based on perceptual wavelet denoising

This paper presents a new speech enhancement system. A psychoacoustic model is incorporated into the wavelet denoising technique to combat different adverse noise conditions. The system is composed of a feed-forward subsystem connected to the perceptual wavelet transform (PWT) processor, a soft-threshold based denoising scheme, and an unvoiced speech enhancement. The noisy speech is first decomposed into critical bands by the PWT from which a set of weights is extracted by the feed-forward subsystem. The average normalized time-frequency energy is used to guide the feedforward subsystem thresholding to reduce the stationary noise while the non-stationary and correlated noises are reduced by an improved wavelet denoising technique with soft-thresholding. Finally, the unvoiced speech enhancement is applied to further improve the intelligibility of the processed speech signal. Simulation results showed that this new system is capable of reducing noise with little speech degradation and the overall performance is superior to several competitive methods in both objective and subjective evaluations.