The Wavelet Analysis To The Pitch Detection Of Speech Signal

The wavelet transform is a new subject developed quickly in the past ten years. Compared with Fourier transform and Gabor transform, the wavelet transform is a part of time-frequency transform, so the message can be obtained from the signals effectively. By means of the fractionized multi-resolution analysis to the signals, many problems unable to be solved by Fourier transform have been solved in this way. This article introduces the theory of continuous wavelet transform and dyadic wavelet transform, and the practical condition of their converse transform. And more, the way of analyse to the object from general to particular is introduced. Based on the decomposition of function space, Mallat established the theory of multi-resolution when he research the image processing, which we can obtain the useful information through the characteristic of multi-scale of quadrature wavelet.The thoery of multi-resolution contacts the wavelet transform with the engineering application, which gives not only simple way of the structure of quadrature wavelet but also the academic basis of fast arithmetic on wavelet transform.Because the speech signal is periodicity at sonant which vocal cords surge in low frequency and similarity to white noises at surd, the pitch can be detected in traditional way through the correlation operation without the speech produce model. But it is not the same with surd or the alternation of sonant and surd. On the base of the acoustics, a whole linear time-varying discrete speech produce model is established through anatomising the factors including track, driving source and lip eradiation the mechanism of speech in this paper, and we draw a conclusion that the time between glottal closure is pitch period. If the wavelet transform is directly implemented in pitch detection, comparing the glottal closure singularity of speech signal with image grey break, we will not obtain the anticipative result. Based on the speech produce model,we find the reason of periodicity disappearance and the extremum number increase by analysing the character of speech signal when the glottal closes. In addition, we analyse the practical problem of Mallat arithmetic and give the corresponding solution. At last, we take the strongpoint and shortcoming of the two kind of ways ,then combine the correlation and wavelet. The experiment result shows the feasibility of the idea.