Codecfake Dataset

Dataset Overview

Codecfake Dataset is a dataset for universal detection of deep‑fake audio, composed of multiple subsets including training, development, and test sets.

Subset Name	Description	Link
Training set (part 1 of 3) & Labels	train_split.zip & train_split.z01 - train_split.z06	Link
Training set (part 2 of 3)	train_split.z07 - train_split.z14	Link
Training set (part 3 of 3)	train_split.z15 - train_split.z19	Link
Development set	dev_split.zip & dev_split.z01 - dev_split.z02	Link
Test set (part 1 of 2)	Codec test: C1.zip - C6.cip & ALM test: A1.zip - A3.zip	Link
Test set (part 2 of 2)	Codec unseen test: C7.zip	Link

The dataset should be organized as follows:

├── Codecfake │ ├── label │ │ └── *.txt │ ├── train │ │ └── *.wav (740,747 samples) │ ├── dev │ │ └── *.wav (92,596 samples) │ ├── test │ │ └── C1 │ │ └── *.wav (26,456 samples) │ │ └── C2 │ │ └── *.wav (26,456 samples) │ │ └── C3 │ │ └── *.wav (26,456 samples) │ │ └── C4 │ │ └── *.wav (26,456 samples) │ │ └── C5 │ │ └── *.wav (26,456 samples) │ │ └── C6 │ │ └── *.wav (26,456 samples) │ │ └── C7 │ │ └── *.wav (145,505 samples) │ │ └── A1 │ │ └── *.wav (8,902 samples) │ │ └── A2 │ │ └── *.wav (8,902 samples) │ │ └── A3 │ │ └── *.wav (99,112 samples)

If you wish to jointly train with the ASVspoof2019 dataset, first download the corresponding training, development, and evaluation sets from the ASVspoof2019 LA Database.

Several pre‑trained models are provided, including Vocoder‑trained ADD, Codec‑trained ADD, and Co‑trained ADD models, stored in the ./pretrained_model directory.

When using this dataset, please cite:

@article{xie2024codecfake, title={The Codecfake Dataset and Countermeasures for the Universally Detection of Deepfake Audio}, author={Xie, Yuankun and Lu, Yi and Fu, Ruibo and Wen, Zhengqi and Wang, Zhiyong and Tao, Jianhua and Qi, Xin and Wang, Xiaopeng and Liu, Yukun and Cheng, Haonan and others}, journal={arXiv preprint arXiv:2405.04880}, year={2024} }