BanglaBeats: A Comprehensive Dataset of Bengali Songs for Music Genre Classification Tasks

This repository contains the codebase for the models and scripts developed to assess the BanglaBeats dataset. This work was titled BanglaBeats: A Comprehensive Dataset of Bengali Songs for Music Genre Classification Tasks and published at the 26th IEEE ICCIT.

Models and Performance

1. CNN Model: Our Convolutional Neural Network (CNN) model achieved exceptional performance, boasting a test accuracy of 88%. This model surpassed all other existing CNN models in this domain.

2. Pre-trained Models: We also explored the effectiveness of pre-trained models, including DistilHubert and Wav2Vec2-Base-960h. These models yielded impressive test accuracies of 83.36% and 84.94%, respectively.

Repository Contents

This repository includes:

• Source code for all models developed, and finetuned in the study.
• Additional scripting files for data preprocessing, model training, evaluation, and testing.

Dataset Links

• Kaggle: BanglaBeats on Kaggle
• Hugging Face: BanglaBeats on Hugging Face

Citation

If you use BanglaBeats in your work, please cite the following paper:

Title: BanglaBeats: A Comprehensive Dataset of Bengali Songs for Music Genre Classification Tasks
Authors: Md. Mehedi Hasan Jibon, Dewan Mahinur Alam, Mohammad Shahidur Rahman
Conference: 2023 26th International Conference on Computer and Information Technology (ICCIT)
DOI: 10.1109/iccit60459.2023.10441288

BibTeX:

@inproceedings{jibon2023banglabeats,
  title={BanglaBeats: A Comprehensive Dataset of Bengali Songs for Music Genre Classification Tasks},
  author={Jibon, Md Mehedi Hasan and Alam, Dewan Mahinur and Rahman, Mohammad Shahidur},
  booktitle={2023 26th International Conference on Computer and Information Technology (ICCIT)},
  pages={1--6},
  year={2023},
  organization={IEEE}
}