Diagnosis Of Speech Disorders In Parkinson’s Disease Based On Energy Direction Features

Speech disorder is one of the incipient symptoms of Parkinson’s disease(PD).The diagnosis of PD based on speech disorders has developed rapidly and become one of the research hotspots recently.Traditional acoustic feature information is single and details are missing.Deep learning features have problems of over-fitting and poor interpretability.This paper proposes a research method for the diagnosis of speech disorders in PD based on the energy direction features to solve the above problems.The specific work is as follows:Firstly,the theoretical basis and related technologies of this paper are introduced.The research direction of speech feature extraction in this paper is proposed on the basis of traditional acoustic features.At the same time,we introduce the classifier and the principle of classification,the speech dataset of PD,the cross-validation method and the evaluation index in this paper.Secondly,presented the research of PD disorder diagnosis based on Empirical Mode Decomposition(EMD)energy direction features.The EMD is proposed to decompose the speech signal to get different order Intrinsic Mode Function(IMF)so as to get more detailed information of the speech signal.Based on the analysis and summary of the existing literature on the diagnosis of speech disorders in PD,it is concluded that there are a large number of speech signal features in the spectrogram.Based on this,it is proposed to extract the feature of the energy change rate of speech signal from the spectrogram,namely the energy direction feature.Finally,the energy direction features of IMF signals were extracted for the classification and diagnosis of the speech signals of PD patients and healthy people.The classification accuracy was 96.65% in the Dataset-Sakar Data and 96.22% in the Dataset-CPPDD.Finally,presented the research of PD disorder diagnosis based on Adaptive Tunable QFactor Wavelet Transform(Adaptive-TQWT)energy direction features.Through the analysis of PD patients and healthy people pronounce with vocal cord vibration mode,we put forward the use of parameters can be set according to the characters of speech signal to TQWT decomposition technology.In order to eliminate the error caused by the uniform setting of parameters,this paper puts forward the Adaptive-TQWT voice signal decomposition method.By extracting the Adaptive-TQWT direction of the decomposed each subband energy features for speech disorders diagnosed PD.The classification accuracy was 96.65% in Dataset-Sakar and 96.25% in Dataset-CPPDD.