Dataset Card for Test-Audio of CMI-Bench

CMI-Bench: A Comprehensive Benchmark for Evaluating Music Instruction Following

🔗 Paper (arXiv) 🧪 Evaluation Toolkit 📊 License: CC BY-NC 4.0

---

Dataset Summary

The CMI-Bench/test-audio dataset provides the complete test split audio files used in the CMI-Bench benchmark. CMI-Bench evaluates the instruction-following capabilities of audio-text large language models (LLMs) on a wide range of Music Information Retrieval (MIR) tasks.

This collection includes test set audio excerpts from 20 public MIR datasets, covering tasks such as genre classification, emotion tagging, music captioning, lyrics transcription, pitch/key estimation, and beat tracking.

⚠️ This release includes only test set audio files for evaluation purposes, in line with usage terms of the source datasets.

---

unzip dataset:

zip -s0 test_data.zip --out test_data_full.zip
unzip test_data_full.zip

or you can use 7z to unzip the dataset:

7z x test_data.zip

Supported Tasks and Sources

---

Intended Use

This dataset is intended for research purposes only, especially for evaluating audio-text models on instruction-following tasks in music understanding.

Please refer to the CMI-Bench GitHub repository for scripts and metrics to run standardized evaluations.

---

License

This dataset is distributed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0). You are free to share and adapt the data for non-commercial purposes, with proper attribution.

Source data is aggregated from multiple public datasets that may carry their own licenses. Please ensure compliance with any additional dataset-specific terms.

---

Citation

Please cite the following paper if you use this dataset:

@misc{ma2025cmibenchcomprehensivebenchmarkevaluating,
      title={CMI-Bench: A Comprehensive Benchmark for Evaluating Music Instruction Following}, 
      author={Yinghao Ma and Siyou Li and Juntao Yu and Emmanouil Benetos and Akira Maezawa},
      year={2025},
      eprint={2506.12285},
      archivePrefix={arXiv},
      primaryClass={eess.AS},
      url={https://arxiv.org/abs/2506.12285}, 
}