Possibilities of Recognizing Depression with Convolutional Networks Applied in Correlation Structure

In the present study, we investigated the possibility of automatic detection of depression. We used convolutional neural networks applied on a special correlation structure that was extracted from speech signal. The quality of life of an individual with depression is greatly influenced by the severity of the depression, which in extreme cases can lead to suicide. On this basis, it is crucial that the disease is recognized at an early stage and that appropriate treatment is provided. This article describes a method that can detect depression through the use of convolutional neural networks by purely spectral features based on speech signal processing. We present how the accuracy of detecting depression changes based on the use of different acoustic-phonetic features and the correlation structure. Using this method, we were able to distinguish between healthy and depressed individuals based on their speech patterns with an accuracy of 84%.