The research focuses on the automation of depression detection through the integration of Convolutional Neural Networks (CNNs) and Transformer-based Pre-Trained Language models. Traditionally, depression assessment relied on comprehensive clinical interviews where psychologists analyzed responses to assess an individual’s mental state. In this model, the replication of this process incorporates three modalities: word context, audio, and video. Through the fusion of these modalities, the prediction of an individual’s mental health status is achieved. The deep learning model encompasses three distinct data modalities, each representing different degrees of depression experienced by the subject. The model assigns appropriate weights to each modality, generating output based on their collective influence. This research contributes to advancing depression monitoring by employing cutting-edge deep learning techniques to analyze multi-modal data, leveraging independent learning synergies for enhanced accuracy.