Depression is a common disease that affects many people in human society. Stress related to emotions and psychology can be reduced by music therapy. As communication technology has expanded and technology has advanced, wearables are becoming more widely used and more comfortable in people’s lives. In this project, a music recommendation system (MRS) is developed based on deep learning methods and electroencephalogram (EEG) signals to recognize the cognitive emotions of the user to enrich the emotional state. An electrode band is used to capture the EEG signals from significant nodes on the temporal, frontal, and parietal regions of the head in a non-invasive manner. These biopotential signals are filtered, preprocessed and the extracted features are presented to deep learning models for detecting the state of emotion. In this work, LSTM, GRU, and hybrid LSTM-GRU models were implemented for the classification of emotion. The hyperparameters of the models were varied and tuned to identify optimal values for proper classification in each model. A comparative analysis was performed based on accuracy, F1 score, precision, and recall and observed that the performance of hybrid LSTM-GRU model was superior. Based on the detected emotion, namely happy, sad, and depressed, choices of music are recommended for the user based on his current mental state. This music therapy is recommended to medical staff as a basic reference for treatment. To maintain a user-friendly environment, an APP is developed in this project for the user to understand the emotion and to select the choice of songs accordingly.