MusicSet Dataset

Overview

MusicSet is constructed from the MTG‑Jamendo dataset by selecting and expanding music audio and adding descriptive text. The dataset contains roughly 150 k 10‑second music‑text pairs.

Data Processing

1. Audio Selection: Choose music audio with at least five tags.
2. Audio Segmentation: Load audio files, extract the middle 80 %, and split into 10‑second segments, discarding non‑melodic parts.
3. Tag Expansion: Use the DeepSeek API to expand multiple tags into full descriptive texts.
4. Data Integration: Combine the generated music‑text pairs with MusicBench and MusicCaps to form the final MusicSet dataset.

Data Format

• Audio Files: Saved as individual WAV files.
• Description Files: Saved as JSON files.

Citation

@article{wei2024melodyneedmusicgeneration,
      title={Melody Is All You Need For Music Generation}, 
      author={Shaopeng Wei and Manzhen Wei and Haoyu Wang and Yu Zhao and Gang Kou},
      year={2024},
      eprint={2409.20196},
      archivePrefix={arXiv},
      primaryClass={cs.SD},
      url={https://arxiv.org/abs/2409.20196}, 
}