Audio detection based on segmented spectrograms

Detecting depression at an early stage is very important in treating it. In this paper, we propose a model for detecting depression using deep learning for detecting depression using patients' speech features. The model is built by establishing the spectral features of the audio using a Resnet34 network that incorporates the SKNet attention mechanism. This paper is assessed using the DAIC-WOZ dataset provided by the AVEC 2016 Depression Analysis Challenge. The experimental findings demonstrate that the model proposed in this paper surpasses the baseline model and achieves superior results in unimodal features.