[ICLR 2026] AudioX-IFcaps: Instruction-Following Audio Caption Dataset

AudioX-IFcaps (Instruction-Following) is a large-scale, high-quality multimodal dataset designed for training unified audio and music generation models. The dataset contains over 7 million samples with fine-grained, structured annotations that enable precise control over audio generation, including sound event categories, counts, temporal ordering, and timestamps.

📊 Dataset Statistics

• General Audio: ~1.3m 10-second video-audio clips
• Music: ~5.7m 10-second video-music clips
• Total Duration: ~16k hours of audio content

📝 Citation

If you use this dataset in your research, please cite:

@article{tian2025audiox,
  title={Audiox: Diffusion transformer for anything-to-audio generation},
  author={Tian, Zeyue and Jin, Yizhu and Liu, Zhaoyang and Yuan, Ruibin and Tan, Xu and Chen, Qifeng and Xue, Wei and Guo, Yike},
  journal={arXiv preprint arXiv:2503.10522},
  year={2025}
}

@inproceedings{tian2025vidmuse,
  title={Vidmuse: A simple video-to-music generation framework with long-short-term modeling},
  author={Tian, Zeyue and Liu, Zhaoyang and Yuan, Ruibin and Pan, Jiahao and Liu, Qifeng and Tan, Xu and Chen, Qifeng and Xue, Wei and Guo, Yike},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={18782--18793},
  year={2025}
}

🔗 Related Resources

• Paper: AudioX: Diffusion Transformer for Anything-to-Audio Generation (Accepted to ICLR 2026)
• Project Page: https://zeyuet.github.io/AudioX/
• Code: GitHub Repository

---

Note: This dataset is part of the AudioX project. For more information, please refer to the paper and project page.