Speech enhancement by perceptual adaptive wavelet de-noising

This thesis work summarizes and compares the existing wavelet de-noising methods. Most popular methods of wavelet transform, adaptive thresholding, and musical noise suppression have been analyzed theoretically and evaluated through Matlab simulation. Based on the above work, a new speech enhancement system using adaptive wavelet de-noising is proposed. Each step of the standard wavelet thresholding is improved by optimized adaptive algorithms. The Quantile based adaptive noise estimate and the posteriori SNR based threshold adjuster are compensatory to each other. The combination of them integrates the advantages of these two approaches and balances the effects of noise removal and speech preservation. In order to improve the final perceptual quality, an innovative musical noise analysis and smoothing algorithm and a Teager Energy Operator based silent segment smoothing module are also introduced into the system. The experimental results have demonstrated the capability of the proposed system in both stationary and non-stationary noise environments