Decoding Research On Deep Learning Of Parkinson’s Disease Speech Disorder

The detection of Parkinson’s disease by means of speech disorder is one of the hot topic in the early diagnosis of Parkinson’s disease.In recent years,the speech disorder detection based on deep learning has developed rapidly,and the detection accuracy is high.Because of the inherent characteristics of deep learning,the speech features automatically learned through the deep learning network are less explicable in medicine.In order to solve this problem,this paper proposes to study the relationship between Mel-Frequency Cepstral Coefficients and speech depth learning feature of Parkinson’s disease(PD).In this paper,the partial least squares method,multitasking Lasso and multitasking elastic network are used to carry out a series of discussions.Firstly,the speech data set is classified by convolution neural network,and the pre-training model is obtained.The pre-training model is used to extract the deep learning features of speech,and the Mel-Frequency Cepstral Coefficients of speech is extracted at the same time.Then the correlation analysis of the above two features is carried out to demonstrate the feasibility of the method proposed in this paper.In this paper,a decoding framework for decoding speech features from Mel-Frequency Cepstral Coefficientss is proposed.Secondly,the partial least square method is used to map the two feature sets of the speech data training set,and the number of components of the partial least squares method is determined by contrast experiments.The partial least squares model obtained from the training set is used to decode the Mel-Frequency Cepstral Coefficientss of the test set speech,and the quality test of the deep learning decoding features of the speech is further carried out,and the partial least square model is analyzed.Thirdly,the multi-task Lasso is used to realize the multivariate mapping between the two feature sets of the speech data training set,and the optimization of the parameters of the multi-task Lasso model is realized by contrast experiments.The optimized multi-task Lasso model is used to decode the Mel-Frequency Cepstral Coefficientss of the test set.The quality test of the deep learning decoding features of the speech is carried out and the multi-task Lasso model is analyzed.Finally,the multi-task elastic network is used to realize the multivariate mapping of the two feature sets of the speech data training set,and the optimization of the parameters of the multi-task elastic network model is accomplished through multi-group contrast experiments.The optimized multi-task elastic network model is used to decode the Mel-Frequency Cepstral Coefficientss of the test set,and the deep learning features of the speech are tested for the quality,and the multi-task elastic network model is carried out.At the same time,We will have a series of discussions about the partial least squares method,multitasking Lasso and multitasking elastic network.