Dataset for AUDETER: A Large-scale Dataset for Deepfake Audio Detection in Open Worlds

Dataset Description

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.

Please refer to our GitHub repository for more information.

Dataset Structure

The dataset is organized into four main corpus collections:

├── common_voice/          # Common Voice corpus
│   ├── tts/              # TTS models
│   │   ├── bark/
│   │   ├── chattts/
│   │   ├── cosyvoice/
│   │   ├── f5_tts/
│   │   ├── fish_speech/
│   │   ├── openvoice/
│   │   ├── sparktts/
│   │   ├── vits/
│   │   ├── xtts/
│   │   ├── yourtts/
│   │   └── zonos/
│   └── vocoders/         # Vocoders
│       ├── bigvgan/
│       ├── bigvsan/
│       ├── full_band_melgan/
│       ├── hifigan/
│       ├── melgan/
│       ├── multi_band_melgan/
│       ├── parallel_wavegan/
│       ├── style_melgan/
│       ├── univnet/
│       └── vocos/
├── in-the-wild/          # In-the-wild corpus
│   ├── tts/              # TTS models
│   └── vocoders/         # Vocoders
├── mls/                  # Multilingual LibriSpeech corpus
│   ├── tts/              # TTS models
│   └── vocoders/         # Vocoders
└── peoples_speech/       # People's Speech corpus
    ├── tts/              # TTS models
    └── vocoders/         # Vocoders