Music recommendation system based on face emotion recognition using Convolution Recurrent Neural Network(CRNN)

Abstract

The emergence of commercial music streaming services, which can be accessed via mobile devices, has greatly enhanced the accessibility of digital music compared to previous periods. Organizing digital music can be a laborious and even overwhelming task, resulting in information overload. Therefore, the creation of an automated music recommender system that can effectively search music collections and offer suitable song recommendations to consumers is quite beneficial. By employing a music recommender system, music providers can predict and deliver appropriate song recommendations to users based on the audible characteristics of previously listened-to music. The aim of our study is to create and execute a music recommender system that leverages audio signal properties to offer recommendations based on similarity. This study leverages a convolutional recurrent neural network (CRNN) to extract features and employs a similarity distance metric to evaluate the similarity of these derived features. The study's findings indicate that consumers have a preference for recommendations that consider music genres rather than those that solely concentrate on similarities. The system's performance was assessed using the FER2023 dataset and compared to the most advanced techniques, resulting in an accuracy of 85%.