BMGC: A Deep Learning Approach to Classify Bengali Music Genres
Moumita Sen Sarma, Avishek Das · 2021
Music genre classification (MGC) is the process of tagging music with their appropriate genres by analyzing music signals or the lyrics. With the accelerated surge in music data repositories, MGC can be extensively used in music recommendation systems, advertisement, and streaming services for systematic and efficient management. However, there have been many works on English music classification using different statistical and machine learning approaches, but there is no notable progress found in the arena of Bengali music. Besides, a few significant works have been found in utilizing Deep Learning (DL) methods to classify different music genres. Bengali music is significantly enriched with its contents and uniqueness. Moreover, the extent and scope of exploring the DL approach in Bengali music ground are still latent. Therefore, Bengali music genre classification is quite a new research area in the Deep learning field. In this work, we have constructed a Bengali Music Genre Classifier (BMGC) to categorize 6 Bengali music genres: ‘Adhunik’, ‘Band’, ‘Hiphop’, ‘Nazrulgeeti’, ‘Lalon’, and ‘Rabindra Sangeet’. We have created a Bengali music genre classification dataset (hereafter named BMGCD) containing 2944 Bengali music clips, and a Gated Recurrent Unit based deep learning model has been developed to predict the music genre from audio signals. Our developed model achieved an accuracy of 80.4% and 80.6% F1-score which surpasses the related existing works.