Parkinson's Disease Auxiliary Diagnosis Based On Speech Feature Extractio

Parkinson’s disease has always brought many difficulties to human beings in their later years,and its prevalence has been increasing in recent years.Because voice disturbance is a typical early symptom of Parkinson’s disease,some researchers have attempted to diagnose the disease based on voice data collected from patients.Although existing methods can provide acceptable results,they are only suitable for some scenarios.In other words,they are neither generative nor robust enough.Therefore,how to determine Parkinson’s disease from voice signals has always been a difficult problem.This paper is an exploration and research on the direction of Parkinson’s disease based on voice signals.The research content of this paper is as follows:(1)Aiming at the problem of Parkinson’s disease speech signal classification,a method of introducing deep learning is proposed to solve it.First,for the preparation of the data set,the method of extracting the spectrogram from the speech signal is introduced in detail,and the Mel filter and the constant Q transform in the speech signal processing are introduced.After processing,the two public Parkinson’s disease speech database was converted into the Parkinson’s disease speech image data set required for the experiment,and then the Xception network and Res Net50 network were used to conduct experiments.The experiment proved that the method of Parkinson’s speech classification based on the spectrogram has certain problems,but the method is feasible.It can realize the recognition of Parkinson’s disease based on speech signals.(2)Aiming at the difficulty of feature extraction in speech classification,study the use of neural network to directly extract features from speech signal data and apply it to Parkinson’s disease speech classification.Use the Vggish model to process speech data,learn the features in the speech data for classification,and perform multiple experiments on different machine learning classification methods to demonstrate the feasibility of the method.And compared with the Parkinson’s disease speech classification of image classification,the performance of this set of methods based on deep learning to extract speech signal features has been improved to a certain extent.(3)Aiming at the problems determined by the Parkinson’s speech classification feature classifier,it is proposed to divide the speech classification of Parkinson’s disease into two modules of speech signal feature extraction module and feature classification module for research.For the feature extraction module,PRAAT acoustic analysis software is used to extract the basic features of the speech signal;and the method of mathematical statistics is used to extract audio features.For the feature classification module,based on the obtained speech features,the optimization algorithm is applied to the search of the classification network to obtain a suitable classifier.Based on the principles of the differential evolution algorithm and the particle swarm optimization algorithm,a method for the classification network model search is constructed.The final experiments have confirmed the practicability of the whole set of methods and the improvement of accuracy in Parkinson’s disease speech classification.