Research And Application Of Depression Recognition Based On Speech Signal

As one of the currently high prevalence mental diseases,depression has many characteristics such as the large number of patients,the long duration,and the harmfulness of results,which make it the biggest challenge in the diagnosis of mental illness.Faced with the fact that the number of people suffering from depression has been increasing year by year,it is the most effective way to improve the recognition rate of depression and make early diagnosis and treatment.According to the clinical medical manifestations of patients with depression,it is found that there is a very obvious difference in the features of speech between depression patients and normal groups,such as lower pitch and slower speech speed.Therefore,this dissertation is based on the speech signal for the research and application of depression recognition.Research has shown that to improve the accuracy of depression recognition,it is usually possible to obtain better recognition results from the extraction of speech features with depressive patients' and the use of different classification methods.This dissertation mainly carries on the experiment of the subject from a variety of algorithms.Based on the two depression speech datasets of AViD-Corpus and DAIC-WOZ,the preprocessing and feature extraction of speech signals in datasets were completed in the early stage,and the model training as well as results analysis of the classification methods were implemented in the later period.In view of the fact that deep learning method currently has a good classification performance in the recognition of depression,this dissertation proposes a method using generative adversarial network combined with convolutional neural network based on speech signals to research of depression recognition in order to achieve better recognition result.Previous research results have shown that the ensemble learning method has a good recognition effect in the experimental research work.Based on this,this dissertation also proposes using generative adversarial network combined with ensemble learning method for the research of depression recognition on the experimental data.This dissertation uses combination of a variety of deep learning methods based on speech signals to achieve accurate identification and classification of depressive patients and normal groups,the statistical analysis data of multiple groups of experimental results show that the method proposed in this dissertation improves the accuracy of depression recognition and have a certain degree of comparability.In the end,this dissertation designed and implemented a depression recognition system based on the Android platform.