| Summary: | 碩士 === 中華技術學院 === 電子工程研究所碩士班 === 96 === Speech signal is the most convenient and shortcut way of intercommunion. However, in the course of the intercommunion, speech signal is disturbed and polluted inevitably by surroundings and transmission mediums. At the same time, it also influences the other transact processing of speech signal, for example, speech identification, speech code and so on. so it is important to denoise for the speech signal. In the past, we enhance the speech signal by the traditional methods, including time-domain, frequency-domain and windowed Fourier transform, but there are shortcomings that limit their application. Wavelet transform is a very popular time-frequency analyzing method developed in the late 1980’s. Wavelet analysis structure with time frequency character and is applied popularly in fact and plays a more important role in digital signal denoising domain.
This paper mainly presents the basic knowledge of speech enhancement and wavelet transform, as well as the studying background and studying progress. Common methods for speech enhancement and wavelet transform was expatiated and studied. On the base of this, the principle and realizing steps of endpoint detection and speech denoising based on wavelet transform was studied, and problems of selection of wavelet base and decomposition level were discussed. A method of speech signal denoising of multi-scale and multi-threshold based on wavelet transform was presented. The simulation results show that this algorithm is valid on white noise conditions especially when that SNR is low. Experimental results show that this algorithm is capable to improve the SNR of the speech and is a nice improving scheme for algorithms which are based on soft and hard threshold.
Key words: wavelet transform; threshold function; signal to noise ratio; white noise
|
|---|
