Machine Learning and Deep Learning Techniques For Genre Classification of Bangla Music
Towkir Ahmed, M. Arfayet Alam, Rakesh Roshan Paul, Tanvir Hasan, Raqeebir Rab · 2022
Music genre classification is extremely important for both music recommendation and acquisition of music data, as well as for music discovery. There have already been a vast amount of researches conducted on the classification of music genres using various machine learning algorithms. Despite the fact that Bangla music is extremely diverse in terms of its own style, there has been little notable work done to date to categorize song genres in Bangla music using machine learning approaches. There are numerous varieties and modes of Bangla music, all of which may be categorised into different classes by their musical compositions. The dataset we use contains six different Bangla music genres. There are several unique attributes for each song which is included in the dataset, including zero crossing value, delta, chroma frequency, spectral roll-off, spectral bandwidth, and many others. Several machine learning models, as well as a deep learning technique, are proposed in this paper for classi-fying Bangla musics into multi-class classification. To train the supervised learning models, we used dimentionality reduction and feature scaling to increase the performance. Finally, our models are evaluated using f'l-score, recall, accuracy and precision. As can be observed, the implemented deep neural network model was able to reach an accuracy of 77.68 percent.