Research And System Implementation Of Speech Detection Algorithm For Parkinson’s Disease

Neurodegenerative diseases are becoming more common,affecting areas of the brain such as movement and speech.The application of artificial intelligence technology in the medical field has made the detection and diagnosis of diseases more accurate.Parkinson’s Disease(PD),as a typical neurodegenerative disease,has some problems such as early diagnosis obstacles,low efficiency of diagnosis methods and subjectivity.Dysarthria is one of the early symptoms observed in PD,and 90% of patients will experience it.Early diagnosis of PD can be timely intervention and treatment to delay the development of the disease.The main research content of this article is as follows.(1)Based on the complex tone system of Chinese,combined with the pronunciation of Chinese PD patients,this paper formulated a series of phonetic tasks including vowels,short sentences,words and numbers by referring to the Chinese Standard Aphasia Checklist of China Rehabilitation Research Center,and constructed a data set of Mandarin PD speech.(2)The machine learning method was used to construct a speech classification model,and the four types of acoustic feature parameters including linear,nonlinear,time-frequency and prosodic parameters were extracted from the collected vowels,words and short sentences.The optimal feature subset was constructed through correlation analysis and significance difference analysis,and the Support Vector Machines were selected.SVM),Multilayer Perceptron(MLP),Random Forest(RF),K-Nearest Neighbors(KNN),Logistic Regression,LR)and Naive Bayes(NB)classifiers to construct classification models.the classification accuracy of words and sentences was higher than vowels.The SVM model achieved the highest classification accuracy for short sentences,and the Area Under the Curve(AUC)was 0.94.The classification accuracy of MLP and SVM models was the highest in word tasks,with an AUC of 0.92.The RF model achieved the highest classification accuracy on vowel signals,with an AUC of 0.79.The results show that the machine learning method is feasible to realize PD speech diagnosis,and the accuracy of model classification depends on feature extraction,feature subset construction and classifier selection.(3)The deep learning method is used to build a speech classification model,and the collected speech signals of unit sounds,vowels,numbers,words and short sentences are converted into spectral graphs,and PD speech spectral data sets are constructed.The Gaussian fuzzy method is used to complete the expansion of data sets.When VGG16 network is selected,the accuracy of the final model on the test set reaches 89.7%,and the AUC reaches 0.975.Then,VGG16 network model was improved,and SVM,RF,MLP,KNN,LR and NB classifiers were used to replace the fully connected layer of the network to complete the classification task.RF model achieved the highest classification accuracy of 93.31%,which was 3.61% higher than the original model.Finally,based on the training neural network classification model,the speech diagnosis system is designed and implemented,which is presented to users in the form of Graphics User Interface.