Study Of Optimization In Parameters Of Pathological Voice

Voice is extremely important in people’s daily life and work, however, voice diseases is inevitably brought to person on account of the negative factors of life and work. Hence, the detection and treatment of pathological voice must be done at the moment. Essentially, in the detection and treatment of pathological voice, the feature parameters are used to do contrastive analysis and to be classified into some voice category. Because the analytical method of feature parameter of pathological voice comes from the identity of voice signal, acoustics analytical method used in voice signal is suitable for the pathological voice. In the passed several decades, researchers find that the boundedness exists in the description of pathological voice by means of traditional acoustic features and come up with a more effective method-nonlinear mechanical technology analysis because of voice signal producing mechanical nonlinear process chaos. Based on former research of this study, the paper mainly describes an optimization in parameters combination of pathological voice feature and the method to abstract nonlinear features, which brings about reference for clinically detecting pathological voice with noise without any damage and real time. This work abstract the effective features information, reduce the expense of space and time raise the rate of recognition of feature parameters, and diagnose the illness in shorter time.While automatically detecting and diagnose pathological voice, only the feature parameters containing enough kinds of information can get a better rate recognition. The rest parameters containing one feature neither represent the feature of the whole voice information nor meet the requirement of real time. However, combined features represent voice information rather comprehensively, giving arise to need to detect voice feature by the set of multi-feature-parameter combination. Besides taking this set into account, the redundance and related information between (and in) features must be removed. The paper presents an approach based on Kernel Principal Component Analysis (KPCA) to combine the feature parameters of pathological voice. The approach optimizes the combination set of acoustic features and nonlinear features parameters, and then recognises the voice by the best set and SVM. The research result shows that the by means of combination set based on KPCA, the highest and average rate of recognition of vowel/a/reach 97.47% and 91.85%,91.39% and 84.15% for the vowel/i/. Compared with the traditional method, the average rates have improved to some extent.At the same time, the nonlinear mechanical feature of the voice effectively differentiates the normal and pathological voice, but takes to much amount of calculation and time, which does harm to process real time. So it is necessary that the paper finds a method to reduce the time to abstract parameters but does not sacrifices the rates of recognition of pathological voice, even has these rates improved. For this purpose, this paper presents a method based on human ear-filter banks to abstract nonlinear features of pathological voice. During the abstraction, the periodical feature vectors, which can optimize the feature are, used to abstract the feature. The research shows that this method sharply shortens the time of abstract, and the parameters acquired from the method gets a better recognition, in which the best is based on the human eras filter bank of Gammachirp.