The timing of intervention is very important in the treatment prognosis of depression disorders. If it is delayed, symptoms can worsen or lead to secondary diseases. Therefore, early diagnosis of depression is very important. Since patients may find i...
The timing of intervention is very important in the treatment prognosis of depression disorders. If it is delayed, symptoms can worsen or lead to secondary diseases. Therefore, early diagnosis of depression is very important. Since patients may find it difficult to visit a psychiatrist, many studies on developing deep learning and machine learning models that can easily detect depression disorders have been preceded under the name of Automatic Depression Detection(ADD). However, such studies still have limitations in that they use professional data that is difficult to obtain, such as blood test results and electroencephalogram.
Accordingly, number of studies have also been conducted to develop ADD models using data more readily available in everyday life, such as voice and video. However, in order for users to trust the results of the model, it needs to provide a reason for judgment. In this work, we developed an explainable deep learning model that can detect depression disorders using patient's voice data.
In this work, we used LSTM and Self-Attention Mechanism to develop explainable ADD models. The model is of academic significance in that it can easily visualize the reason of detection using speech data. Experimental results for evaluating the performance of the model are presented later in the paper.