Deep Learning based Song Recommendation System for Emotions of the Listener

People usually listen to music according to their moods and emotions. With the rapid growth of digital music libraries manual selection of a song or a list of songs is cumbersome. Music or song classification and recommendation have risen in popularity. In this paper, a system that can recommend songs according to the emotions of listeners is developed. Emotions are classified with the help of the Viola-Jones algorithm and a Long Short Term Memory Network (LSTM) neural network. Viola Jones algorithm is used to detect the face in an image in the first stage and an LSTM is used to classify the emotions into 4 categories. A second network is trained to classify the songs using another LSTM and according to the emotion of the given image, a song or a list of songs is recommended. The performance in terms of accuracy of 96.7% is achieved.