Study On The Novel Method Of Pathological Voice Diagnosis Based On Acoustics, Wavelet Entropy And Auto-Regressive Model

The computerized pathological voice evaluation and diagnosis is a new field for the early therapy, illness detection and medical treatment informatization technology. Since its objective, rapid and general used character it is becoming the important and assistant measurement to prevent the ealy diagnose some fatal diseases. The novel method of pathological voice evaluation and diagnosis mentioned in this paper is a tool of early detection and diagnosis of illness wich is to be concerned. This area falls back on the bases of acoustics and anatomia theory by means of computer and morden digital signal processing technique to analyze the sound sample data in time or spectrum domain and classify them in multidimensional features space.In the research the database is supplied by Masachusetts Eye and Ear Infirmary. The major research is including the extractions of 19 aouctics, 5 wavelet entropy and 13 Auto-Regressive model features straightly from the voice samples. The support vector machine is employed as the classifier in bi-classification of healthy and pathology voice with the different feature group of above the three. And then the further task is hyperfunction and fold damage illness discrimination. At last dependin on the final experments result, the hyperfunction and fold damaged illness specialized features are founded among these groups' features by using the orthogonal experiment design. The experments effects show that acoustics and Auto-Regressive model goroup's features are appearances obvious in healthy and pathology division. The accuracies are 89.7% and 87.8% respectively. Thought the Auto-Regessive model features are first used, it not only well done in pathological voice detection but also has a best obviuos performenance and much better than the other two groups features in pathological voice identification with accuracy of 87.3%. The 1st to 5th, 8th to 9th and 11th AR model coefficients are choosen and determined to have specialized characteristics.The innovations of this thesis are the followings:i. The first combination using of acoustics, wavelet entropy and Auto-Regressive model features in research of pathological voice diagnosis. It is the great expandation work to the using features set and benifical to illness realy stage diagnosis.ii. For the first time to do the dieseases indentification between the hyperfunction and fold damaged illness. The specialized features which have the particular probobility to indentify the hyperfunction or fold damaged illness. This is an adventage of sublclassification in diagnosis precisly.iii. First use of support vector machine as the classifier for pattern recognation, which improve the precision of the classification algorithem and accuracy of the effects.