Speech Singnal De_noising Research Based On Wavelet Transformation

In daily life,communication between people cannot be separated from speech.Speech is also an important medium for transmitting information.It is a time-varying and non-stationary random signal.People will be influenced by the ambient noise when they communicate with each other,and will also be subjected to noise interference in the process of speech communication.When the external environment noise is similar to the speech spectrum,it is very difficult to remove the noise mixed in the speech signal.It is very important for speech signal processing to analyze the non-stationary speech signal by wavelet transform.In this thesis,a new threshold function de-noising method with better denoising effect is proposed by using the characteristics of wavelet multiresolution analysis and the study of wavelet threshold denoising algorithm.Specific work as follows:1.The principle of Fourier transform and short time Fourier transform is analyzed,and then the wavelet transform and wavelet analysis are extended.The wavelet threshold denoising algorithm based on wavelet transform is emphatically analyzed,and the wavelet decomposition level,threshold,wavelet base and threshold function involved in the wavelet threshold denoising algorithm are deeply analyzed and studied.2.In the traditional wavelet threshold denoising algorithm,the speech after the hard threshold de-noising algorithm will oscillate,and the speech distortion after the soft threshold de-noising is large.A new improved wavelet threshold function is presented in this thesis.The new threshold function algorithm considers the attenuation of the modulus of the noise wavelet transform conforms to the exponent.By adjusting the parameters,the traditional threshold function is avoided directly when the wavelet coefficients are less than the threshold value,thus effectively enhancing the denoising effect.3.In order to further enhance the denoising effect of noisy speech with low SNR,the two step denoising method combining Calman filtering and improved wavelet threshold method is proposed in this thesis.Experiments show that this method can improve speech signal output SNR and reduce the degree of distortion,which is an effective enhancement algorithm.