A Study on Music Genre Classification And Lyrics Generation of Bangla Songs Dataset

This study is divided into two crucial sections. We first gathered Bangla song names and lyrics using web crawling and a dataset that was only slightly annotated. A comprehensive dataset was then thoroughly annotated across five different genres. Utilizing transformer models to classify the genre of Bangla songs was our main goal. Noteworthy accomplishments are the 94% accuracy and F1-score of csebuetnlp/banglabert. Additionally, we created a text generation system using ritog/bangla-gpt2 and flax-community/gpt2-bengali. This method allows for the development of scenario-based music material, expanding the creative possibilities for the entertainment business, by choosing the genre with the highest F1 score from the classification findings.