Research And Realization Of Speech Enhancement Method Based On Wavelet Transform

Speech enhancement is a method of extracting speech information from noisy speech signal, which is extensively used in speech recognition, speech coding, speech communication and other fields. Hence, speech enhancement plays an important role in speech signal processing. This dissertation deeply studied the enhancement methods of speech signal which is polluted by additive noise. The main works and innovations are as follows: (1) The basic principle of several speech enhancement methods are studied in this dissertation, such as harmonic enhancement method, the enhancement method based on speech generation model, spectral subtract method, Wiener filter method ,etc. The strongpoint and shortcoming of the above methods are also compared with each other in this dissertation. (2) Different characteristics of speech signal wavelet analysis coefficients and noise wavelet analysis coefficients are analyzed, and the basic theory of wavelet transform, basic principle and methods of wavelet denoising are discussed. The characteristics comparison have been made between different threshold estimation methods and between different threshold quantization methods when they are used to eliminate noise fromnoisy signal, and the effectiveness of wavelet denoising is verified by using experiment. Wavelet field speech enhancement method based on threshold is studied, and an improved speech enhancement method of wavelet packet analysis is introduced. The simulation results indicate that: this method can effectively enhance the noisy speech signal that was polluted by additive white noise and additive airplane noise. Namely, the signal noise ratio(SNR) is increased, and the hearing easiness of speech is improved.(3) A speech enhancement module based on TMS320VC5416 fixed point DSP is designed, and wavelet field speech enhancement algorithm based on threshold is realized by this module. The design process of hardware circuit of this module and the design thought of the algorithm are discussed in detail. Noisy speech signal collected in the noisy environment are processed. The experiment results indicate that this module can satisfy the requirement of speech enhancement, so the effectiveness of algorithm is verified.