Speech Detection of Depression Based on Multi-mlp

Depression is one of the most prevalent mental illnesses on a global scale. In view of the inefficiency of current depression screening methods. This paper proposes a depression detection model based on a deep model hybrid architecture to assist doctors in diagnosing depression. 157 Chinese subjects were investigated in this study. It is worth noting that we propose a word reading experiment to make the subject’s emotional change rapidly. We extract the Low-level audio features to find out the different emotional change in the process of reading different parts of speech words. We use convolutional neural network to extract deep spectrum features and Multi-mlp to detect depression. The experimental results show that the accuracy rate of speech depression recognition reaches 82.70%, which can effectively assist doctors in diagnosing depression.