biboamy / music-repro

Music Genre Classification with Reprogramming

Paper Link

• Low-Resource Music Genre Classification with Advanced Neural Model Reprogramming.
• Yun-Ning Hung, Chao-Han Huck Yang, Pin-Yu Chen, and Alexander Lerch

Codebase

1. Download GTZAN dataset: https://www.kaggle.com/datasets/andradaolteanu/gtzan-dataset-music-genre-classification?resource=download

2. unzip and move file:

unzip archive.zip
mv Data/genres_original/* music-repro/data/

3. Install Dependencies

pip3 install -r requirement.txt

4. Pull pre-trained models

git lfs fetch --all

5. run experiment (skip to "7" for inference only)

python3 main.py --reprog_front uni_noise

python3 main.py --reprog_front condi

python3 main.py --reprog_front skip

6. Visit "demo.ipynb" for inference only demo

AST Reference

@inproceedings{gong21b_interspeech,
  author={Yuan Gong and Yu-An Chung and James Glass},
  title={{AST: Audio Spectrogram Transformer}},
  year=2021,
  booktitle={Proc. Interspeech 2021},
  pages={571--575},
  doi={10.21437/Interspeech.2021-698}
}

The ast code used in this repo comes from the original repo

Citing Music Reprogramming

@inproceedings{hung2023low,
  title={Low-Resource Music Genre Classification with Cross-Modal Neural Model Reprogramming},
  author={Hung, Yun-Ning and Yang, Chao-Han Huck and Chen, Pin-Yu and Lerch, Alexander},
  booktitle={Proc. of ICASSP 2023},
  pages={1--5},
  year={2023},
  organization={IEEE}
}

Bug Fix

If you encounter the following errors "batch response: This repository is over its data quota. Account responsible for LFS...", Please download the model from here Google Drive