Study On Speech Disorders Of Parkinson’s Disease Based On Local Information Analysis In Transform Domain

Speech disorder is an early symptom of Parkinson’s disease,and the detection of Parkinson’s disease based on speech disorder is developing rapidly.In the last few years,the speech diagnosis of Parkinson’s disease on time-frequency transform domain is one of the hot topics in speech research of Parkinson’s disease.Aiming at the analysis of the speech signal in time-frequency transform domain of Parkinson’s disease,this paper conducts a series of studies from the perspective of local energy change.Firstly,the speech signal is converted into two transform domain energy signals by shorttime Fourier transformation and Mel frequency transformation.In the research of transform domain,by analyzing the direction information of local energy distribution in transform domain,a speech disorder analysis method based on local direction statistics is proposed.First,the angle relationship between the energy point and its surrounding energy point is calculated,and the statistical characteristics of local direction are obtained by statistical angle relation.Then dimension reduction is used to transform the local direction statistical features,and the reduced features are classified by classifier.Finally,this method is evaluated and verified by multiple groups of comparative experiments.Secondly,based on the analysis of amplitude difference information of energy distribution in transform domain,a research method based on local difference statistics is proposed by analyzing the amplitude difference information of energy distribution in the transform domain to diagnose speech disorders in Parkinson’s disease.First,the local energy D-value information of the transform domain is calculated,and the statistical features of local D-value are obtained through statistical information of local difference.Then the statistical features are de-dimensionalized,and the statistical features of local D-value after the reduction are classified by the classifier.Finally,the statistical features of D-value are evaluated according to the results of sets of comparative experiments.Finally,combining the direction of energy transformation and the amplitude difference in the transform domain,this paper proposes an extraction method of statistical features based on local gradient.Firstly,the gradient and angle of each energy point in the transform domain are calculated.And the gradient of local energy points is counted by the angle to obtained the statistical features of local gradient.Finally,the dimension of local gradient statistical features is reduced and the reduced gradient statistical features are classified by classifier.In the evaluation stage,the performance of the local gradient statistical features is evaluated through multiple groups of comparative experiments,and the three local statistical features are compared and analyzed.Compared with other literatures,the result explains the advanced nature of this method.