Pitch Period Extraction Of A-P Squeezing Using WAC Method

Pathological voice is caused by the change of sound sick rational, not completelyclosure of the glottis or the irregular vibration of vocal cords, introducing the noise toaudible system. With the aggravation of the symptoms, the noise component will continueto increase. Acoustic detection method can effectively diagnose the type and severity ofvoice diseases by studying various acoustic parameters in depth.On the basic research of pathological voice, aiming at A-P squeezing of pathologicalvoice, this paper propose one method combining the wavelet transformation and theautocorrelation method (WAC) to extract the pitch period. It also uses7kinds of classifierto recognize the pathological voice. The main works are as following:The pitch frequency of pathological voice is unstable and sometimes it is disappear. Inorder to solve this problem, this paper takes use of the wavelet transformation to filter thenon-sharp change information and noise, enhancing the sharp change information ofinstant glottal closure. It is superior to the traditional short-time Fourier transform,providing a powerful tool for non-stationary signal analysis and also the possibility formore accurate voice signal pitch period estimation.When the voice signal is polluted by noise, as the decreases of signal-to-noise ratio,the pitch period error obtained using wavelet transform method is increasing. In order toconquer this problem, this paper takes use of one more accurate pitch period detectionmethod, combining the wavelet transformation and the autocorrelation method. It regardsthe transformed signal as the signal to be processed, calculating the autocorrelationfunction. As for the maxima point exceeding the certain threshold limitation, we shouldmeasure the interval between the adjacent two maxima, thereby obtaining the pitch period of the speech signal.Then, on the basis of accurately extracting the pitch period, according to the acousticparameters related to the pitch period, this paper takes use of several classifiers torecognize the A-P squeezing of pathological voice with different degree. The experimentalresults show that the recognition rate of WAC method can reach89.83%.At last, this paper raises the shortcomings of this method and the problems thathaven’t been solved, and gives the direction of further study and improving.