Data sets and benchmarks for audio deepfake detection
| Name | Dateset size | Languages | Attack types | Evaluation metrics | Year |
|---|---|---|---|---|---|
| ASVspoof 2015 (Wu et al., 2015) | 260k+ | English | VC, statistical TTS | EER | 2015 |
| ASVspoof 2017 (Kinnunen et al., 2017) | 18k+ | English | Replay attacks | EER | 2017 |
| ReMASC (Reynolds et al., 2019) | 55k+ | English | Replay, manipulation | EER, ROC-AUC | 2019 |
| ASVspoof 2019 (Todisco et al., 2019) | 360k+ | English | Logical access (VC/TTS), physical access (replay) | EER, t-DCF | 2019 |
| FoR (Reimao and Tzerpos, 2019) | 198k | English | TTS, VC, replay | Accuracy, EER | 2019 |
| WaveFake (Müller et al., 2021) | 105k-118k | English | Neural TTS, VC (multiple architectures) | EER, accuracy | 2021 |
| FakeAVCeleb (Khalid et al., 2021) | 500 celebrities | English | Audio-visual deepfakes (TTS + face manipulation) | Accuracy, EER | 2021 |
| ASVspoof 2021 (Yamagishi et al., 2021) | 500k+ | English | Logical access, physical access, deepfake | EER, t-DCF | 2021 |
| ADD (Yi et al., 2022) | 500k+ | English, Mandarin | TTS, VC, hybrid | EER, accuracy | 2022 |
| LibriSeVoc (Sun et al., 2023) | 90k+ | English | Vocoder artifacts | EER, robustness | 2023 |
| ASVspoof 5 (Wang et al., 2025) | 1M+ | English | Crowdsourced deepfakes, adversarial attacks | EER, t-DCF, mindcf, Cllr | 2025 |
| Name | Dateset size | Languages | Attack types | Evaluation metrics | Year |
|---|---|---|---|---|---|
| ASVspoof 2015 ( | 260k+ | English | VC, statistical | 2015 | |
| ASVspoof 2017 ( | 18k+ | English | Replay attacks | 2017 | |
| ReMASC ( | 55k+ | English | Replay, manipulation | EER, ROC-AUC | 2019 |
| ASVspoof 2019 ( | 360k+ | English | Logical access (VC/ | EER, t-DCF | 2019 |
| FoR ( | 198k | English | TTS, VC, replay | Accuracy, | 2019 |
| WaveFake ( | 105k-118k | English | Neural TTS, | EER, accuracy | 2021 |
| FakeAVCeleb ( | 500 celebrities | English | Audio-visual deepfakes ( | Accuracy, | 2021 |
| ASVspoof 2021 ( | 500k+ | English | Logical access, physical access, deepfake | EER, t-DCF | 2021 |
| 500k+ | English, Mandarin | TTS, VC, hybrid | EER, accuracy | 2022 | |
| LibriSeVoc ( | 90k+ | English | Vocoder artifacts | EER, robustness | 2023 |
| ASVspoof 5 ( | 1M+ | English | Crowdsourced deepfakes, adversarial attacks | EER, t-DCF, mindcf, Cllr | 2025 |
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.