Speech signal is the carrier of the speech information, and also is the most nature, convenient and shortcut way of intercommunication. However in most cases, speech signal is disturbed and polluted inevitably so that we can't catch the explicit meaning of the speech signal.Wavelet transform theory is one of the modern analysis methods. Wavelet threshold de-noising is a de-noising technique based on the threshold of the wavelet coefficients. In this paper,several approaches of speech enhancement are introduced briefly. Wavelet threshold de-noising is the main part in this paper . Two aspects improvement is presented in this paper. In one aspect,two new threshold function is presented based on the analysis of the traditional soft-threshold, hard-threshold functions and the threshold function presented by L.Breiman . In the another aspect, a new idea about threshold estimation which is affected by the idea of spectral subtraction is proposed. Experiment results demonstrate that this method is effective in aspect of remaining geometrical characteristics of signal and improving the signal-to-noise ratio (SNR). |