Music Recommendation System based on Emotion Detection using Image Processing and Deep Networks

Abstract

There are several techniques in which a user can decide the music they want to listen to from a variety of collections. Playing music based on the mood of the user is an application of deep learning which is introduced to the listeners. This can be done by figuring out the user's facial expression which depicts their mood. Expressions talk a lot more about humans than words do. Due to increased numbers of songs being produced every day, it is becoming difficult for the users to figure out the song they would want to listen to. Our Recommendation System aims to solve this problem by providing users a list of songs directly into their personalised playlist based on the mood. The aim is to build a music recommendation system using image processing which is based on neural networks. The novelty lies in the fact that we plan to leverage a set of model architecture which has been developed by the scientist at Meta. The Multitask Cascaded Convolution Neural Network(MTCNN) and FaceNet Architecture have been used for face detection and recognition through the embedding generated. We then run a Convolution Neural Network Model to predict the emotion. Finally a classifier is used to recommend music based on their emotion from a spotify dataset which consists of almost 6 hundred thousand songs. The user can set the number of songs he/she wants to be recommended. We will also add the recommended songs back to user playlist.