We introduce AUDETER (AUdio DEepfake TEst Range), a large-scale, highly diverse deepfake audio dataset for comprehensive evaluation and robust development of generalised models for deepfake audio detection. It consists of over 4,500 hours of synthetic audio generated by 11 recent TTS models and 10 vocoders with a broad range of TTS/vocoder patterns, totalling 3 million audio clips.
| Collection | Subset | Partition | Patterns | # Audio / Patt. | Total Hrs |
|---|---|---|---|---|---|
| TTS | In-the-Wild | Bona-fide | 15 | 19,784 | 311.5 |
| Common Voice | Val | 15 | 16,372 | 275.0 | |
| Test | 15 | 16,372 | 265.3 | ||
| People Speech | Val | 15 | 18,622 | 493.5 | |
| Test | 15 | 34,898 | 909.4 | ||
| MLS | Test | 15 | 3,807 | 212.7 | |
| Test | 15 | 3,769 | 209.1 | ||
| Vocoder | In-the-Wild | Bona-fide | 10 | 19,784 | 207.6 |
| Common Voice | Val | 10 | 16,372 | 266.7 | |
| Test | 10 | 16,372 | 264.8 | ||
| People Speech | Val | 10 | 18,622 | 331.7 | |
| Test | 10 | 34,898 | 598.1 | ||
| MLS | Dev | 10 | 3,807 | 156.7 | |
| Test | 10 | 3,769 | 154.9 |
Table 2: The structure of the AUDETER dataset.
Due to the size of our dataset, we are uploading the dataset in progress. Thanks for your understanding.
You can view the upload in progress on Hugging Face.
If you use AUDETER in your research, please consider the paper and giving us a star🌟!
@article{wang2025audeter,
title={AUDETER: A Large-scale Dataset for Deepfake Audio Detection in Open Worlds},
author={Wang, Qizhou and Huang, Hanxun and Pang, Guansong and Erfani, Sarah and Leckie, Christopher},
journal={arXiv preprint arXiv:2509.04345},
year={2025},
url={https://arxiv.org/abs/2509.04345}
}