Datasets:
metadata
license: cc-by-nc-4.0
tags:
- benchmark
- music-generation
- evaluation
size_categories:
- 1G<n<10G
CTIM-Bench: Benchmark for Traditional Chinese Music Generation
CTIM-Bench is a curated evaluation dataset consisting of 500 expert-validated high-fidelity samples, specifically designed to benchmark music generation models on traditional Chinese instruments. It serves as the test set for the CTIM-Gen project.
The dataset features 42.6% minority ethnic instruments (e.g., Matouqin, Hulusi), reflecting the rich diversity of Chinese musical culture.
⚖️ Data Source & Usage Policy
The audio data in this benchmark was primarily collected from China Music Network (www.china1901.com), with a small amount supplemented from Bilibili.
Important: This dataset is provided solely for academic research purposes in the field of AI music generation. Commercial use is strictly prohibited. By downloading or using this dataset, you agree to these terms.
Contents
- audio/: Ground truth audio files (30s clips).
- test_prompts.json: Metadata containing prompts and file references.
- Instruments: Guqin, Guzheng, Matouqin, Hulusi, Pipa, Erhu, Yangqin.