| Depression is a kind of mood disorder disease with clinical manifestations of significant depression,most or complete loss of pleasure and interest,long attack duration and long-term repetition.Without intervention,patients will have serious behaviors such as suicide.However,with the rapid increase in the number of patients with depression,there is no corresponding increase in the number of hospitals and doctors.Therefore,computer technology to assist in the diagnosis of depression came into being,which is the general trend of the development of the times.Computer technology to assist in the diagnosis of depression can detect some details that doctors may ignore.It has incomparable advantages in capturing patients’ micro expressions,extracting patients’ audio features,analyzing patients’ words and carrying micro information.This thesis studies and designs a depression detection system based on speech and text features.The main work is as follows:1.An ATDD database is built,and the database data is intercepted and marked by professional doctors.The sound data processed by doctors are preprocessed,and the corresponding sound spectrum is obtained as the main feature of the voice detection system.The voice is transformed into text for the text feature detection system.2.In the aspect of speech based depression detection network,the channel weight iterative update mechanism is introduced to improve the traditional residual network model.In the aspect of depression detection network based on text features,the local and global text analysis methods are compared,and the global text analysis method based on Doc2vec and TextCNN with better effect is selected.3.For each patient,it is proposed to fuse the depression detection results based on speech and text features,and output the results of multimodal fusion.The experimental results fully show that the depression detection method based on speech and text features explored in this thesis is feasible and has high accuracy.At the same time,after multimodal fusion,the accuracy of classification will be improved.It is also confirmed that using multimodal network is better than using single-mode voice or text feature network.The depression detection system based on voice and text features designed in this thesis can be used as an auxiliary means for doctors to detect depression.While reducing the pressure of diagnosis and treatment and greatly improving the efficiency,it is also of great help to improve the diagnosis rate. |
PDF Full Text Request