Pith. sign in

REVIEW 2 major objections 4 minor 1 cited by

Circumventing shortcuts in audio-visual deepfake detection datasets with unsupervised learning

T0 review · 2 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Fake videos in two major audio-video deepfake benchmarks begin with a telltale silence that lets a trivial classifier score over 98% AUC—and removing that silence collapses several published detectors.

desk verdict A real, previously missed shortcut: fake clips in two standard audio-visual deepfake datasets start with a brief silence, and the paper shows prior benchmark results partly ride on it. read the letter →

arxiv 2412.00175 v3 pith:HI7YB5LF submitted 2024-11-29 cs.CV cs.LGcs.SDeess.ASeess.IV

Stefan Smeu , Dragos-Alexandru Boldisor , Dan Oneata , Elisabeta Oneata This is my paper · ORCID
classification cs.CVcs.LGcs.SDeess.ASeess.IV
keywords deepfakedetectionaudio-visualleadingsilencedatasetbiasspuriousfeaturesunsupervisedlearningself-supervisedalignmentAV-HuBERT
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that two widely used audio-video deepfake benchmarks—FakeAVCeleb and AV-Deepfake1M—carry an unintended shortcut: fake videos begin with about 25–30 ms of silence, while real videos start with noise. A trivial classifier that measures only the duration of this leading silence separates real from fake samples with over 98% AUC on both datasets. The authors show that several published audio and audio-visual detectors have been exploiting this feature, and that removing the silence degrades them sharply, for example dropping the MDS method from 99.2% to 54.9% AUC on AV-Deepfake1M. To avoid such dataset-specific biases, they propose training only on real videos: AVH-Align learns an audio-visual alignment score from self-supervised features and uses misalignment to flag fakes, without ever seeing a fake during training. If this holds, previous benchmark results partly reflect an artifact, and the field needs evaluation protocols and learning paradigms that are robust to such shortcuts.

What carries the argument

The central mechanism is the leading-silence duration: the time until the audio magnitude first exceeds a small threshold ($5\cdot10^{-4}$), a feature that alone yields AUC above 98% on both datasets. The proposed countermeasure, AVH-Align, uses a pretrained AV-HuBERT model to extract frame-level audio and visual features, then trains a four-layer MLP alignment network $\Phi$ on real videos only, using an InfoNCE-style contrastive loss that maximizes the match between each audio frame and its corresponding visual frame while downweighting neighbouring frames. At inference, the negative of the per-frame alignment score, pooled with log-sum-exp, gives the fakeness score. The paper also uses a trimming protocol—discarding the leading silence rounded up to the video's frame period—to evaluate whether a method has latched onto the artifact.

What would settle it

Collect or create deepfakes using the same generators (e.g., TalkLip, VITS) but with audio and video streams length-matched before packaging and check whether fake samples still begin with 25–30 ms of silence; if they do, the silence is intrinsic to the generation process, not the dataset construction, and the artifact interpretation fails.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the leading silence is a spurious feature that pervades two major audio-visual deepfake datasets and inflates the measured performance of existing detectors. On the paper's own terms, the finding is that fake videos in FakeAVCeleb and AV-Deepfake1M start with a brief moment of quiet audio—around 25–30 ms—whereas real videos begin with noise, so a classifier built solely on this feature reaches over 98% AUC. The authors further show that state-of-the-art audio and audio-visual models rely on this silence: after trimming it, RawNet2 loses 11.8 points of AUC on AV-Deepfake1M and MDS collapses from 99.2% to 54.9% AUC. They propose that the remedy is to abandon supervised training on fake data altogether and instead learn on real videos only, and they demonstrate AVH-Align, which aligns self-supervised AV-HuBERT audio-visual features with a frame-level contrastive network and scores fakeness by the degree of audio-visual misalignment. AVH-Align is unaffected by the silence shortcut and achieves the highest reported AUC (85.24%) among methods that do not exploit the bias on the official AV-Deepfake1M test set.

Load-bearing premise

The leading silence is a dataset-construction artifact rather than a real property of deepfakes; if genuine manipulated videos commonly start with silence, trimming it would discard a true signal and the evaluation would be misleading.

Editorial extensions

If this is right

  • Reported scores on FakeAVCeleb and AV-Deepfake1M should be treated as untrustworthy unless accompanied by results on silence-trimmed data.
  • Existing deep-learning detectors that trained on these datasets may generalize much worse than their benchmarks suggest; a silence probe is a cheap robustness check.
  • Training on real data only, as AVH-Align does, removes the incentive to exploit dataset asymmetries and remains competitive with supervised approaches.
  • Audio-only methods hold up well after trimming, indicating the audio stream carries strong, underused signal for deepfake detection.
  • The official AV-Deepfake1M test set shares the silence bias, so leaderboard entries without a bias check should be interpreted cautiously.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next step is to apply the leading-silence probe to other audio-visual datasets, since the same packaging pipeline (generated audio appended to real video) could introduce the artifact elsewhere.
  • The alignment-score signal that AVH-Align produces is frame-level, and the paper's 77.7% frame-level AUC suggests the approach could be extended from detection to localization of manipulated segments.
  • The trim-and-remeasure protocol could become a general evaluation discipline: for any suspected shortcut feature, compare performance with and without it to quantify reliance.
  • If the silence arises because generated audio is shorter than the video, dataset curators could fix it by length-matching rather than trimming, which would also preserve synchronization.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper reports a leading-silence bias in two widely used audio-visual deepfake datasets, FakeAVCeleb and AV-Deepfake1M: in the fully fake (FVFA) subsets, fake videos begin with roughly 25-30 ms of silence, and a simple threshold-based silence-duration classifier reaches above 98% AUC. The authors then show that trimming this leading silence degrades some supervised models, most notably MDS on AV-Deepfake1M, and propose AVH-Align, an unsupervised method trained only on real videos by aligning AV-HuBERT audio and video features. AVH-Align is reported to be robust to trimming and to achieve 85.24% AUC on the official AV-Deepfake1M test set, surpassing the visual-only baselines reported there. The paper concludes that prior supervised results partly reflect this spurious cue and that unsupervised real-data training is a useful evaluation and modeling paradigm.

Significance. If the bias is as general as claimed, the paper makes a valuable benchmarking contribution: it identifies a concrete, easily checked artifact, shows that it inflates several published results, and proposes a simple robustness test (trimming the leading silence). The silence analysis is strengthened by threshold-robustness experiments and by an independent maximum-amplitude measure, and the trimming experiments convincingly show that some supervised models rely on the cue. The official test-set evaluation of AVH-Align is a strong point, since it avoids self-reported validation-set scores. The proposed method is simple and reproducible, with code released. The main limitations are that the headline claim is overbroad relative to the actual experimental scope, and that the causal interpretation of the silence as a dataset-construction artifact is only speculative.

major comments (2)
  1. [Abstract and Section 3.2] The headline claim that FakeAVCeleb and AV-Deepfake1M suffer from a leading-silence artifact, with a simple classifier reaching over 98% AUC, is computed only on the RVRA versus FVFA subsets, as stated in Section 3.2 and Table 1. On FakeAVCeleb's FVRA split (fake video, real audio), the leading-silence AUC is only 62.4%, and Section 3.2 reports that LAV-DF, AVLips, and DFDC do not exhibit the bias. The abstract's unqualified sentence "Fake videos start with a very brief moment of silence" is therefore overbroad. This is load-bearing because the claim that prior audio-visual results "partly reflect this artifact" is only established for fakes with synthesized audio. Please qualify the claims to the RVFA/FVFA fully-fake or synthesized-audio subsets, and state explicitly which fake splits are used in the full-dataset experiments in Tables 2 and 3.
  2. [Section 3.2, "Why do fake samples have a leading silence?"] The paper's only causal explanation for the leading silence is speculative: "we speculate that this happens when the audio may be slightly shorter than the video counterpart." The interpretation of the silence as a dataset-construction artifact is load-bearing for the conclusion that trimming removes a spurious cue and that prior results are overestimated. If the TTS or lip-sync tools used to create the fakes (SV2TTS, VITS, YourTTS, TalkLip) intrinsically produce a short leading silence, then the silence is a genuine property of fakes produced by those tools, and the trimmed evaluation would discard a valid discriminative signal. The indirect evidence (absence in LAV-DF, AVLips, DFDC, and weaker bias on FVRA) does not establish the causal origin. Please either provide direct evidence about the generator outputs before dataset assembly, or rephrase the conclusions conditionally, making clear that "spurious" means "not shared across deepfake generation pipelines" rather than "not caused by the fake-generation tools."
minor comments (4)
  1. [Section 4.1, Eq. (3)] Equation (3) writes p(v_i|a_j) but the surrounding text and Eq. (2) use p(v_i|a_i); the subscript in Eq. (3) should presumably be a_i.
  2. [Section 4.2] The text says the AUC of AVH-Align/sup "degrades by 16.9%" when going from 100.0 to 83.1 AUC; since AUC is already a percentage, this should be expressed as 16.9 percentage points to avoid ambiguity with a relative drop.
  3. [Section 5] There is a typo in the discussion of Rajan et al.: "reconstructring" should be "reconstructing".
  4. [Section 4.3, Table 2] The text says AVH-Align is "not impacted" by leading silence, but the AV-Deepfake1M AUC drops from 85.9 to 83.5 after trimming; please describe this as a small but nonzero effect rather than no effect.

Circularity Check

0 steps flagged · score 0.0 of 10

Leading-silence finding is empirical and externally evaluated; no load-bearing self-citation or fitted prediction.

full rationale

The paper's central claims are (a) a leading-silence artifact exists in FakeAVCeleb and AV-Deepfake1M, and (b) the proposed AVH-Align method is robust to it. Claim (a) is established by direct waveform measurement: the silence duration is defined by an amplitude threshold, the threshold is swept over a grid with stable AUC (Fig. 3, 'What counts as silence?'), and the same bias is independently confirmed on the official held-out AV-Deepfake1M test server (Table 3). No model is fitted to produce the silence classifier; it is simply a ranking by measured silence duration. The subsequent degradation of RawNet2, MDS, and AVAD under trimming is an external measurement using published code or checkpoints, not a prediction derived from the paper's own fitted parameters. Claim (b) is also non-circular: AVH-Align is trained only on real audio-video pairs with an InfoNCE-style alignment loss, and its inference score is the negative per-frame alignment; no constant is tuned using the reported test accuracy. The only self-citations, [40] and [51], appear in contextual remarks about self-supervised audio features and localization ability and are not load-bearing. The speculation in Sec. 3.2 about the cause of the silence is explicitly labeled as speculation and represents a validity limitation rather than circular reasoning. The evaluation against external benchmarks and the robustness analyses make the derivation self-contained.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claims rest on a few assumptions about the nature of the silence artifact, the validity of the trimming procedure, and the transferability of pretrained AV-HuBERT features. No new entities or forces are introduced. The main free parameters are the silence threshold and the neighborhood size.

free parameters (2)
  • Silence amplitude threshold tau = 5e-4 (relative magnitude)
    Used to define where non-silent audio begins when computing the leading silence. The authors show the resulting AUC is stable across a grid of values, so it is not a unique or particularly sensitive choice, but it is still a manually chosen threshold.
  • Temporal neighborhood size N(i) = 30 frames
    The alignment loss in Eq. 2 treats the 30 neighboring frames around each time step as negatives in a softmax. No ablation reports the sensitivity to this choice, so the method may depend on it.
assumptions (4)
  • domain assumption The leading silence in fake videos is a dataset artifact rather than an intrinsic property of manipulated media.
    The trimming experiments and the framing of the shortcut depend on this. The paper only speculates about the cause and provides indirect evidence from other datasets, which is reasonable but not conclusive.
  • domain assumption Trimming the leading silence rounded to the nearest frame boundary preserves audio-video synchronization.
    The trimmed evaluation in Sec. 3.3 assumes that cutting the initial silence at a frame boundary does not introduce new desynchronizations that would artificially lower the performance of audio-visual models.
  • domain assumption AV-HuBERT features encode audio and video in a shared space where cross-modal alignment is informative for manipulation detection.
    The method relies on pretrained AV-HuBERT features whose alignment scores are assumed to transfer to new videos, even when the feature extractor was pretrained on the same source dataset as the test data.
  • standard math InfoNCE-style contrastive loss with a softmax over a local neighborhood provides a valid training signal for alignment.
    Eq. 2-3 are a standard contrastive formulation for matching audio and video frames, widely used in representation learning.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Circumventing shortcuts in audio-visual deepfake detection datasets with unsupervised learning." pith.science (2026). https://pith.science/paper/HI7YB5LF

@misc{pith2026241200175,
  author       = {Pith},
  title        = {Pith review of: Circumventing shortcuts in audio-visual deepfake detection datasets with unsupervised learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HI7YB5LF}},
  note         = {Machine review of arXiv:2412.00175}
}
read the original abstract

Good datasets are essential for developing and benchmarking any machine learning system. Their importance is even more extreme for safety critical applications such as deepfake detection - the focus of this paper. Here we reveal that two of the most widely used audio-video deepfake datasets suffer from a previously unidentified spurious feature: the leading silence. Fake videos start with a very brief moment of silence and based on this feature alone, we can separate the real and fake samples almost perfectly. As such, previous audio-only and audio-video models exploit the presence of silence in the fake videos and consequently perform worse when the leading silence is removed. To circumvent latching on such unwanted artifact and possibly other unrevealed ones we propose a shift from supervised to unsupervised learning by training models exclusively on real data. We show that by aligning self-supervised audio-video representations we remove the risk of relying on dataset-specific biases and improve robustness in deepfake detection.

Figures

Figures reproduced from arXiv: 2412.00175 by the authors.

Figure 1
Figure 1. Audio-visual deepfake detection datasets have a silence [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Left: The impact of the silence threshold on the leading [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Overview of the AVH-Align method. A: We use the pretrained AV-HuBERT model to extract self-supervised features which we further align with a learnable network Φ. Note that we use a single AV-HuBERT model, but make two forward passes to obtain audio-only and video-only features (instead of a single set of multimodal features). B: At training we maximize the alignment score Φii, between the audio features ai at time s… view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Per frame fakeness probabilities for AVH-Align and AVH-Align/sup on AV-Deepfake1M. AVH-Align/sup always marks the first [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Unmasking Synthetic Realities in Generative AI: A Comprehensive Review of Adversarially Robust Deepfake Detection Systems

    cs.CR 2025-07 conditional novelty 3.0 of 10

    A systematic review of deepfake detection finds a pervasive lack of adversarial robustness evaluation across all modalities and calls for resilient, modality-agnostic detectors.

Reference graph

Works this paper leans on

60 extracted references · 48 canonical work pages · cited by 1 Pith paper

  1. [1]

    MesoNet: A compact facial video forgery detection network

    Darius Afchar, Vincent Nozick, Junichi Yamagishi, and Isao Echizen. MesoNet: A compact facial video forgery detection network. InWIFS, page 1–7, 2018. 2, 7

    work page 2018

  2. [2]

    Lost in translation: Lip- sync deepfake detection from audio-video mismatch

    Matyas Bohacek and Hany Farid. Lost in translation: Lip- sync deepfake detection from audio-video mismatch. In CVPR, pages 4315–4323, 2024. 2

    work page 2024

  3. [3]

    Is synthetic voice detection research going into the right direction? InCVPRW, pages 71–80, 2022

    Stefano Borz `ı, Oliver Giudice, Filippo Stanco, and Dario Al- legra. Is synthetic voice detection research going into the right direction? InCVPRW, pages 71–80, 2022. 1

    work page 2022

  4. [4]

    Glitch in the ma- trix: A large scale benchmark for content driven audio-visual forgery detection and localization.Comput

    Zhixi Cai, Shreya Ghosh, Abhinav Dhall, Tom Gedeon, Kalin Stefanov, and Munawar Hayat. Glitch in the ma- trix: A large scale benchmark for content driven audio-visual forgery detection and localization.Comput. Vis. Image Un- derst., 236:103818, 2023. 1, 2, 4

    work page 2023

  5. [5]

    MARLIN: Masked autoencoder for facial video rep- resentation learning, 2023

    Zhixi Cai, Shreya Ghosh, Kalin Stefanov, Abhinav Dhall, Jianfei Cai, Hamid Rezatofighi, Reza Haffari, and Munawar Hayat. MARLIN: Masked autoencoder for facial video rep- resentation learning, 2023. 2, 7

    work page 2023

  6. [6]

    A V-Deepfake1M: A large-scale LLM-driven audio-visual deepfake dataset, 2024

    Zhixi Cai, Shreya Ghosh, Aman Pankaj Adatia, Munawar Hayat, Abhinav Dhall, Tom Gedeon, and Kalin Stefanov. A V-Deepfake1M: A large-scale LLM-driven audio-visual deepfake dataset, 2024. 1, 2, 7

    work page 2024

  7. [7]

    Edresson Casanova, Julian Weber, Christopher Dane Shulby, Arnaldo Cˆandido J ´unior, Eren G ¨olge, and Moacir A. Ponti. YourTTS: Towards zero-shot multi-speaker TTS and zero- shot voice conversion for everyone. InICML, pages 2709– 2720, 2022. 3

    work page 2022

  8. [8]

    What makes fake images detectable? understanding prop- erties that generalize

    Lucy Chai, David Bau, Ser-Nam Lim, and Phillip Isola. What makes fake images detectable? understanding prop- erties that generalize. InECCV, pages 103–120, 2020. 1

    work page 2020

Show all 60 references
  1. [9]

    Xception: Deep learning with depthwise separable convolutions

    Franc ¸ois Chollet. Xception: Deep learning with depthwise separable convolutions. InCVPR, pages 1800–1807, 2017. 7

    2017

  2. [10]

    Not made for each other-audio- visual dissonance-based deepfake detection and localization

    Komal Chugh, Parul Gupta, Abhinav Dhall, and Ra- manathan Subramanian. Not made for each other-audio- visual dissonance-based deepfake detection and localization. InACMM, pages 439–447, 2020. 4, 7

    2020

  3. [11]

    J. S. Chung, A. Nagrani, and A. Zisserman. V oxCeleb2: Deep speaker recognition. InInterspeech, 2018. 2

    2018

  4. [12]

    Combining EfficientNet and vision transformers for video deepfake detection

    Davide Alessandro Coccomini, Nicola Messina, Claudio Gennaro, and Fabrizio Falchi. Combining EfficientNet and vision transformers for video deepfake detection. InICIAP, pages 219–229, 2022. 7

    2022

  5. [13]

    Raising the bar of AI-generated image detection with CLIP

    Davide Cozzolino, Giovanni Poggi, Riccardo Corvi, Matthias Nießner, and Luisa Verdoliva. Raising the bar of AI-generated image detection with CLIP. InCVPR, pages 4356–4366, 2024. 4

    2024

  6. [14]

    Zero-shot detection of AI-generated im- ages

    Davide Cozzolino, Giovanni Poggi, Matthias Nießner, and Luisa Verdoliva. Zero-shot detection of AI-generated im- ages. InECCV, pages 54–72, 2024. 2

    2024

  7. [15]

    Real time speech enhancement in the waveform domain

    Alexandre D ´efossez, Gabriel Synnaeve, and Yossi Adi. Real time speech enhancement in the waveform domain. InInter- speech, pages 3291–3295, 2020. 3

    2020

  8. [16]

    The deepfake detection challenge (dfdc) dataset.arXiv preprint arXiv:2006.07397, 2020

    Brian Dolhansky, Joanna Bitton, Ben Pflaum, Jikuo Lu, Russ Howes, Menglin Wang, and Cristian Canton Ferrer. The deepfake detection challenge (dfdc) dataset.arXiv preprint arXiv:2006.07397, 2020. 4

    2006 arXiv

  9. [17]

    Self- supervised video forensics by audio-visual anomaly detec- tion

    Chao Feng, Ziyang Chen, and Andrew Owens. Self- supervised video forensics by audio-visual anomaly detec- tion. InCVPR, pages 10491–10503, 2023. 2, 4, 5, 6

    2023

  10. [18]

    Lips don’t lie: A generalisable and robust approach to face forgery detection

    Alexandros Haliassos, Konstantinos V ougioukas, Stavros Petridis, and Maja Pantic. Lips don’t lie: A generalisable and robust approach to face forgery detection. InCVPR, pages 5039–5049, 2021. 2

    2021

  11. [19]

    Leveraging real talking faces via self- supervision for robust forgery detection

    Alexandros Haliassos, Rodrigo Mira, Stavros Petridis, and Maja Pantic. Leveraging real talking faces via self- supervision for robust forgery detection. InCVPR, pages 14950–14962, 2022. 2

    2022

  12. [20]

    A VTENet: Audio-visual transformer-based ensemble network exploiting multiple ex- perts for video deepfake detection.CoRR, abs/2310.13103,

    Ammarah Hashmi, Sahibzada Adil Shahzad, Chia-Wen Lin, Yu Tsao, and Hsin-Min Wang. A VTENet: Audio-visual transformer-based ensemble network exploiting multiple ex- perts for video deepfake detection.CoRR, abs/2310.13103,

    arXiv

  13. [21]

    Implicit identity driven deepfake face swapping detection

    Baojin Huang, Zhongyuan Wang, Jifan Yang, Jiaxin Ai, Qin Zou, Qian Wang, and Dengpan Ye. Implicit identity driven deepfake face swapping detection. InCVPR, pages 4490– 4499, 2023. 2

    2023

  14. [22]

    Weiss, Quan Wang, Jonathan Shen, Fei Ren, Zhifeng Chen, Patrick Nguyen, Ruoming Pang, Ig- nacio L ´opez-Moreno, and Yonghui Wu

    Ye Jia, Yu Zhang, Ron J. Weiss, Quan Wang, Jonathan Shen, Fei Ren, Zhifeng Chen, Patrick Nguyen, Ruoming Pang, Ig- nacio L ´opez-Moreno, and Yonghui Wu. Transfer learning from speaker verification to multispeaker text-to-speech syn- thesis. InNeurIPS, pages 4485–4495, 2018. 2

    2018

  15. [23]

    Hasam Khalid, Shahroz Tariq, Minha Kim, and Simon S. Woo. FakeA VCeleb: A novel audio-video multimodal deep- fake dataset. InNeurIPS Datasets and Benchmarks, 2021. 1, 2

    2021

  16. [24]

    Conditional variational autoencoder with adversarial learning for end-to- end text-to-speech

    Jaehyeon Kim, Jungil Kong, and Juhee Son. Conditional variational autoencoder with adversarial learning for end-to- end text-to-speech. InICML, pages 5530–5540, 2021. 3

    2021

  17. [25]

    DeepFakes: A new threat to face recognition? Assessment and detection.CoRR, abs/1812.08685, 2018

    Pavel Korshunov and S ´ebastien Marcel. DeepFakes: A new threat to face recognition? Assessment and detection.CoRR, abs/1812.08685, 2018. 2

    2018 arXiv

  18. [26]

    Fast face-swap using convolutional neural networks

    Iryna Korshunova, Wenzhe Shi, Joni Dambre, and Lucas Theis. Fast face-swap using convolutional neural networks. InICCV, pages 3697–3705, 2017. 2

    2017

  19. [27]

    DiMoDif: Dis- course modality-information differentiation for audio-visual deepfake detection and localization.CoRR, abs/2411.10193,

    Christos Koutlis and Symeon Papadopoulos. DiMoDif: Dis- course modality-information differentiation for audio-visual deepfake detection and localization.CoRR, abs/2411.10193,

    arXiv

  20. [28]

    KoDF: A large-scale Korean deepfake detection dataset

    Patrick Kwon, Jaeseong You, Gyuhyeon Nam, Sungwoo Park, and Gyeongsu Chae. KoDF: A large-scale Korean deepfake detection dataset. InICCV, pages 10744–10753,

  21. [29]

    Zero-shot fake video detection by audio-visual consistency.CoRR, abs/2406.07854, 2024

    Xiaolou Li, Zehua Liu, Chen Chen, Lantian Li, Li Guo, and Dong Wang. Zero-shot fake video detection by audio-visual consistency.CoRR, abs/2406.07854, 2024. 2

    2024 arXiv

  22. [30]

    SpeechForensics: Audio-visual speech representation learn- ing for face forgery detection

    Yachao Liang, Min Yu, Gang Li, Jianguo Jiang, Boquan Li, Feng Yu, Ning Zhang, Xiang Meng, and Weiqing Huang. SpeechForensics: Audio-visual speech representation learn- ing for face forgery detection. InNeurIPS, 2024. 6

    2024

  23. [31]

    Lips are lying: Spotting the temporal inconsistency between audio and visual in lip- syncing deepfakes

    Weifeng Liu, Tianyi She, Jiawei Liu, Boheng Li, Dongyu Yao, Ziyou Liang, and Run Wang. Lips are lying: Spotting the temporal inconsistency between audio and visual in lip- syncing deepfakes. InNeurIPS, pages 91131–91155, 2024. 4, 8

    2024

  24. [32]

    When synthetic traces hide real content: Analysis of stable diffu- sion image laundering.CoRR, abs/2407.10736, 2024

    Sara Mandelli, Paolo Bestagini, and Stefano Tubaro. When synthetic traces hide real content: Analysis of stable diffu- sion image laundering.CoRR, abs/2407.10736, 2024. 8

    2024 arXiv

  25. [33]

    TGIF: Text-guided inpainting forgery dataset.CoRR, abs/2407.11566, 2024

    Hannes Mareen, Dimitrios Karageorgiou, Glenn Van Wallendael, Peter Lambert, and Symeon Papadopoulos. TGIF: Text-guided inpainting forgery dataset.CoRR, abs/2407.11566, 2024. 8

    2024 arXiv

  26. [34]

    Do GANs leave artificial fingerprints? In MIPR, pages 506–511, 2019

    Francesco Marra, Diego Gragnaniello, Luisa Verdoliva, and Giovanni Poggi. Do GANs leave artificial fingerprints? In MIPR, pages 506–511, 2019. 2

    2019

  27. [35]

    M ¨uller, Franziska Dieckmann, Pavel Czempin, Roman Canals, and Konstantin B ¨ottinger

    Nicolas M. M ¨uller, Franziska Dieckmann, Pavel Czempin, Roman Canals, and Konstantin B ¨ottinger. Speech is silver, silence is golden: What do ASVspoof-trained models really learn?CoRR, abs/2106.12914, 2021. 1, 4

    2021 arXiv

  28. [36]

    M ¨uller, Piotr Kawa, Wei Herng Choong, Edres- son Casanova, Eren G ¨olge, Thorsten M ¨uller, Piotr Syga, Philip Sperl, and Konstantin B¨ottinger

    Nicolas M. M ¨uller, Piotr Kawa, Wei Herng Choong, Edres- son Casanova, Eren G ¨olge, Thorsten M ¨uller, Piotr Syga, Philip Sperl, and Konstantin B¨ottinger. MLAAD: The multi- language audio anti-spoofing dataset. InIJCNN, pages 1–7,

  29. [37]

    FSGAN: Subject agnostic face swapping and reenactment

    Yuval Nirkin, Yosi Keller, and Tal Hassner. FSGAN: Subject agnostic face swapping and reenactment. InICCV, pages 7183–7192, 2019. 2

    2019

  30. [38]

    Towards uni- versal fake image detectors that generalize across generative models

    Utkarsh Ojha, Yuheng Li, and Yong Jae Lee. Towards uni- versal fake image detectors that generalize across generative models. InCVPR, pages 24480–24489, 2023. 2, 4

    2023

  31. [39]

    A VFF: Audio-visual feature fusion for video deepfake detection

    Trevine Oorloff, Surya Koppisetti, Nicol `o Bonettini, Di- vyaraj Solanki, Ben Colman, Yaser Yacoob, Ali Shahriyari, and Gaurav Bharaj. A VFF: Audio-visual feature fusion for video deepfake detection. InCVPR, pages 27102–27112,

  32. [40]

    Towards generalisable and cali- brated audio deepfake detection with self-supervised repre- sentations

    Octavian Pascu, Adriana Stan, Dan Oneata, Elisabeta Oneata, and Horia Cucu. Towards generalisable and cali- brated audio deepfake detection with self-supervised repre- sentations. InInterspeech, pages 4828–4832, 2024. 4

    2024

  33. [41]

    Training-free deepfake voice recognition by leveraging large-scale pre-trained models

    Alessandro Pianese, Davide Cozzolino, Giovanni Poggi, and Luisa Verdoliva. Training-free deepfake voice recognition by leveraging large-scale pre-trained models. InACM Work- shop on Information Hiding and Multimedia Security, page 289–294, 2024. 2, 4

    2024

  34. [42]

    Nambood- iri, and C.V

    K R Prajwal, Rudrabha Mukhopadhyay, Vinay P. Nambood- iri, and C.V . Jawahar. A lip sync expert is all you need for speech to lip generation in the wild. InACMM, 2020. 2

    2020

  35. [43]

    On the effectiveness of dataset alignment for fake image detection.CoRR, abs/2410.11835, 2024

    Anirudh Sundara Rajan, Utkarsh Ojha, Jedidiah Schloesser, and Yong Jae Lee. On the effectiveness of dataset alignment for fake image detection.CoRR, abs/2410.11835, 2024. 1, 8

    2024 arXiv

  36. [44]

    Detecting deep- fakes without seeing any.CoRR, abs/2311.01458, 2023

    Tal Reiss, Bar Cavia, and Yedid Hoshen. Detecting deep- fakes without seeing any.CoRR, abs/2311.01458, 2023. 2

    2023 arXiv

  37. [45]

    AEROB- LADE: Training-free detection of latent diffusion images us- ing autoencoder reconstruction error

    Jonas Ricker, Denis Lukovnikov, and Asja Fischer. AEROB- LADE: Training-free detection of latent diffusion images us- ing autoencoder reconstruction error. InCVPR, pages 9130– 9140, 2024. 2

    2024

  38. [46]

    FaceForen- sics++: Learning to detect manipulated facial images

    Andreas Rossler, Davide Cozzolino, Luisa Verdoliva, Chris- tian Riess, Justus Thies, and Matthias Niessner. FaceForen- sics++: Learning to detect manipulated facial images. In ICCV, 2019. 1

    2019

  39. [47]

    Hosler, Paolo Bestagini, Matthew C

    Davide Salvi, Brian C. Hosler, Paolo Bestagini, Matthew C. Stamm, and Stefano Tubaro. TIMIT-TTS: A text-to-speech dataset for multimodal synthetic media detection.IEEE Ac- cess, 11:50851–50866, 2023. 2

    2023

  40. [48]

    A V-Lip-Sync+: Lever- aging A V-HuBERT to exploit multimodal inconsistency for video deepfake detection.CoRR, abs/2311.02733, 2023

    Sahibzada Adil Shahzad, Ammarah Hashmi, Yan-Tsung Peng, Yu Tsao, and Hsin-Min Wang. A V-Lip-Sync+: Lever- aging A V-HuBERT to exploit multimodal inconsistency for video deepfake detection.CoRR, abs/2311.02733, 2023. 2

    2023

  41. [49]

    Learning audio-visual speech representation by masked multimodal cluster prediction

    Bowen Shi, Wei-Ning Hsu, Kushal Lakhotia, and Abdelrah- man Mohamed. Learning audio-visual speech representation by masked multimodal cluster prediction. InICLR, 2022. 2, 5, 8

    2022

  42. [50]

    Detecting deep- fakes with self-blended images

    Kaede Shiohara and Toshihiko Yamasaki. Detecting deep- fakes with self-blended images. InCVPR, pages 18699– 18708, 2022. 2

    2022

  43. [51]

    DeCLIP: Decoding CLIP representations for deepfake localization

    Stefan Smeu, Elisabeta Oneata, and Dan Oneata. DeCLIP: Decoding CLIP representations for deepfake localization. CoRR, abs/2409.08849, 2024. 8

    2024 arXiv

  44. [52]

    Lip reading sentences in the wild

    Joon Son Chung, Andrew Senior, Oriol Vinyals, and Andrew Zisserman. Lip reading sentences in the wild. InProceed- ings of the IEEE conference on computer vision and pattern recognition, pages 6447–6456, 2017. 4

    2017

  45. [53]

    End-to-end anti-spoofing with RawNet2

    Hemlata Tak, Jose Patino, Massimiliano Todisco, Andreas Nautsch, Nicholas Evans, and Anthony Larcher. End-to-end anti-spoofing with RawNet2. InICASSP, pages 6369–6373,

  46. [54]

    Repre- sentation learning with contrastive predictive coding.CoRR, abs/1807.03748, 2018

    A ¨aron van den Oord, Yazhe Li, and Oriol Vinyals. Repre- sentation learning with contrastive predictive coding.CoRR, abs/1807.03748, 2018. 5

    2018 arXiv

  47. [55]

    Tan, and Haizhou Li

    Jiadong Wang, Xinyuan Qian, Malu Zhang, Robby T. Tan, and Haizhou Li. Seeing what you said: Talking face genera- tion guided by a lip reading expert. InCVPR, pages 14653– 14662, 2023. 3

    2023

  48. [56]

    Xin Wang, Junichi Yamagishi, Massimiliano Todisco, H´ector Delgado, Andreas Nautsch, Nicholas W. D. Evans, Md. Sahidullah, Ville Vestman, Tomi Kinnunen, Kong Aik Lee, Lauri Juvela, Paavo Alku, Yu-Huai Peng, Hsin-Te Hwang, Yu Tsao, Hsin-Min Wang, S ´ebastien Le Maguer, Markus B...

    2019

  49. [57]

    DF40: Toward next-generation deepfake detection.CoRR, abs/2406.13495,

    Zhiyuan Yan, Taiping Yao, Shen Chen, Yandan Zhao, Xinghe Fu, Junwei Zhu, Donghao Luo, Li Yuan, Chengjie Wang, Shouhong Ding, and Yunsheng Wu. DF40: Toward next-generation deepfake detection.CoRR, abs/2406.13495,

    arXiv

  50. [58]

    10 A V oiD-DF: Audio-visual joint learning for detecting deep- fake.IEEE Trans

    Wenyuan Yang, Xiaoyu Zhou, Zhikai Chen, Bofei Guo, Zhongjie Ba, Zhihua Xia, Xiaochun Cao, and Kui Ren. 10 A V oiD-DF: Audio-visual joint learning for detecting deep- fake.IEEE Trans. Inf. Forensics Secur., 18:2015–2029, 2023. 2

    2015

  51. [59]

    Attributing fake images to GANs: Learning and analyzing GAN fingerprints

    Ning Yu, Larry S Davis, and Mario Fritz. Attributing fake images to GANs: Learning and analyzing GAN fingerprints. InCVPR, pages 7556–7566, 2019. 2

    2019

  52. [60]

    Exploring temporal coherence for more general video face forgery detection

    Yinglin Zheng, Jianmin Bao, Dong Chen, Ming Zeng, and Fang Wen. Exploring temporal coherence for more general video face forgery detection. InICCV, pages 15024–15034,

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.