The Research Of Extracting Pathological Voice's Characteristics Based On HHT And Recognition

The recognition of pathological voice can make significant meaning to carry out voice examination painless and scatheless technique. Pathological voice recognition must depend on extracting effective characteristic parameter and reasonable recognition method. Whereas, conventional acoustic parameters were extracted mainly base on short stability in speech and then get them by short-time-Foerier transform. Therefor, in this paper we adopt a new signal processing method: Hilbert-Huang Transform (HHT),to extracte a new character of pathological voice: A-f standard deviation parameter. Then we use HMM based on information state change whit time recognition method to validate A-f standard deviation parameter is more resultful.Firstly,we debate and study some key issue in EMD.Moreover, make use of Matlab to achieve EMD and its application.By data simulating analysis, The result indicates that HHT has more excellent time-frequency precision. At last ,we also analyzed what is short and needed to improve in after use.On the basis of reading up correlative information, from the change of energy and frequence ,especially the disorder and perturbation parameter are prominent high in pathological voice,and combined with voice's nonlinear and non-stationary characteristic, by utilizing the character of HHT's excellent time-frequency precision and the filter banks speciality of EMD, we complete the algorithm of how to extracte the newpathological voice's character besides programing。In order to testify A-f standard deviation parameter's validity,the author have designed speech recognition system that based on DHMM . We also affirm the best parameters in the recognition system by experiment.,what include choosing the code's capacitance in VQ , the strcture type and states'transfer matrix etc.Then recognize MFCC and A-f standard deviation parameter respectively. Recogniion result proves the validity of A-f standard deviation parameter. We also discuss further into why the new parameter is more suitable for denoting inherence mechanism of pathological voice by analyzing the two characteristic extracting algorithmic process.Lastly, the article summarize the mainly work ,point out what is short in the method we use now and data collection,give some suggestion on futher task.