The Analysis And Application Of Nonlinear Chaos Theory In Voice Signal Processing For Patients With Stroke

Stroke is a type of disease of high incidence and high mortality rate. In prediction of stroke and the rehabilitation process of patients with stroke, there are no good objective evaluation methods to evaluate the state of patients, only through clinical experience of doctors. This paper combine the physiological characteristics of the process of human vocal with the method of nonlinear dynamic to analyze the voice, extract relevant feature to analyze the state of the broken brain. Here choosing the stroke people to research. In this way, this paper try to find the feature that could measure the state of our brain objectively.In this paper, a method of diagnosis of stroke patients were analyzed and studied and finally achieved the classification of stroke patients and healthy people through voice analysis. The paper mainly includes four aspects (acquisition of stoke patients’voice signal, the analysis of stroke patients’voice signal, construction of feature and classification). The following findings have been achieved.1) A method of study the state of brain through voice analysis has been proposed which is supported by information about the neural mechanisms of human vocal and the brain image. Then this paper introduces the source of voice samples and the syllable of the voice.2) The paper proposed a method based on chaotic characteristic of voice time series and using the nonlinear dynamic method to analyze the state of the brain. Firstly, this paper uses the mutual information method to calculate the time delay and the CAO method to calculate the embedded dimension to reconstruct space and attractor. Then using the small data algorithm to get the first max Lyapunov index. All the mentioned chaotic features prove the chaotic characteristic of the voice time series.3) A feature vector construction method based on improved surrogate data method is proposed for the first time. This method combines the surrogate data method with the correlation dimension to get a new feature-normalized sigma variability. This new feature describes the distinct between the stroke patients’voice and healthy people’s voice which is better than correlation dimension. 4) Pattern recognition is carried out with the feature vector. Firstly, this paper uses the nonlinear method to analyze all the samples which include the voice of stroke people and that of the healthy people and calculates the first minimal value of the mutual information, the correlation dimension, the first max Lyapunov index and the normalized sigma of the signals. The author draws the comparison chart and table of voice samples to vividly describe the difference of the two kind of voice signal. Then use the K nearest neighbor classification method to classify the voice signals. The results shows that the new feature-normalized sigma variability raises the accuracy of the classification and the nonlinear features are possible to distinguish between stroke patients and healthy people. This result also provides direction and basement of the research for using voice the measure of brain states.