Research Of Modeling For Chinese Finals With MP Based On Sine Dictionary

Speech modeling is the basis of speech compression and voice recognition and theuniversal problem of speech signal processing. The problem of Chinese finals modeling is thepart of speech modeling, so Chinese finals modeling has the important significance ofresearch. In this paper, according to acoustic features of dullness for Chinese finals, andChinese finals have obvious harmonic structure, using the matching pursuit method based onsine dictionary to complete Chinese finals modeling. Chinese final is decomposed into a set ofparticular basis functions, and then according to the parameters of basis functions, we canconstruct the approximation of original signal. Specific research work is as follows:First of all, we studied Chinese finals modeling method based on exponential sine atoms.Chinese finals are mainly showed as the characteristics of dullness, and the spectrum ofdullness has the obvious harmonic structure, sinusoidal time-frequency atoms have thecharacteristics of harmonic, so we can use MP method based on exponential sine atoms inChinese finals modeling. First of all, we studied the structure and parameters of exponentialsine atoms, and then we decompose and reconstruct Chinese finals through MP method basedon exponential sine atoms. In the simulation experiment, we contrasted the modeling methodof exponential sine atoms with the modeling method of Gabor atoms, proved the Chinesefinals modeling of exponential sinusoidal atoms is more effective and superior in terms of thenumber of atoms, signal similarity and residual energy.Second, studied Chinese finals sparse representation based on the method oftime-frequency atoms clustering into time-frequency molecular. To the problem of atomicredundancy and large amount of calculation in Chinese final representation with matchingpursuit algorithm, Chinese finals sparse representation based on the method of time-frequencyatoms clustering into time-frequency molecular is proposed. Through the analysis of twoparameters, time shift and frequency, we narrow the search scope of atoms, and reduce thecomputation. And then through fuzzy clustering algorithm, we cluster a bunch oftime-frequency atoms into a time-frequency molecular, and get signal spares representationwith time-frequency molecular.Finally, modeling Chinese finals with the method combined wavelet transform andorthogonal matching pursuit. According to the characteristic of that the frequency of Chinesefinals is mainly concentrated in the low frequency part, we adopted the method of wavelettransform to decomposed Chinese finals’ frequency into low frequency part and highfrequency part, and then used the method of orthogonal matching pursuit to modeling the lowfrequency part of Chinese finals. In the simulation experiment, we contrast the method ofWOMP with MP method and OMP method, verified the method of WOMP can modelingChinese finals efficiently.