Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 86 of the 86 references it cites were found in a registry or an index; 23 of the 43 sources its claims cite were checked for support.

AV-Deepfake1M: A Large-Scale LLM-Driven Audio-Visual Deepfake Dataset

Open access
Cai, Zhixi, Ghosh, Shreya, Adatia, Aman Pankaj, Hayat, Munawar, Dhall, Abhinav, Gedeon, Tom, Stefanov, Kalin · 2024 · Proceedings of the 32nd ACM International Conference on Multimedia

Abstract

Introduces an audio-visual deepfake localization dataset with more than 1M videos and content-driven video, audio, and audio-visual manipulations for more than 2K subjects.

Overview

CoChat Summary

TL;DR

AV-Deepfake1M is a large audiovisual deepfake dataset built with an LLM-driven, three-stage pipeline for transcript, audio, and video manipulation. The paper argues that it is more diverse and harder than prior datasets, and its benchmarks show substantial performance drops for existing detection/localization methods.

Key findings

  • The dataset contains 1,886 hours of audio-visual data from 2,068 unique subjects and 1,146,760 videos total, including 286,721 real and 860,039 fake videos.
  • Compared with LAV-DF, the paper reports much higher generated-media quality: audio quality for AV-Deepfake1M has SECS 0.991, SNR 9.39, FAD 0.088 overall, and visual quality has PSNR 39.49, SSIM 0.977, FID 0.49 overall.
  • Human detection is substantially harder on AV-Deepfake1M than on LAV-DF: user-study accuracy drops from 84.03 to 68.64, AP@0.1 from 36.80 to 15.32, AP@0.5 from 14.17 to 1.92, and AR@1 from 10.04 to 2.54.
  • Existing methods perform much worse on AV-Deepfake1M than on LAV-DF; for example, temporal localization AP@0.5 drops from 79.15 to 37.37 for BA-TFD, from 96.30 to 44.42 for BA-TFD+, and from 98.83 to 51.64 for UMMAFormer.

Methods

  • The paper constructs AV-Deepfake1M from VoxCeleb2 videos using a three-stage pipeline: transcript manipulation, audio generation, and video generation.
  • Transcript manipulation uses ChatGPT via LangChain to perform replace, delete, and insert operations; the number of modifications is set by M = ceil(t / 10), where *t* is video length in seconds.
  • Audio is generated with VITS for some subjects and YourTTS for others; video is generated with TalkLip. The dataset is split into 1,657 train subjects, 411 test subjects, and validation sampled from train; benchmarks evaluate localization and detection with multiple SOTA methods and metrics such as AP/AR and AUC/Acc.

Limitations

  • The paper explicitly states that, like other deepfake datasets, AV-Deepfake1M has a misbalance in the number of fake and real videos.
  • Evident from the text: the test set is restricted to VITS-based identity-dependent audio samples, so not all generation settings appear in test evaluation.
  • Evident from the text: the dataset and benchmarks are oriented to content-driven manipulations on faces/audio-visual speech, so the scope is narrower than general deepfake types.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3664647.3680795

Cites

86 of 86 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

8 Valid · 5 Not valid · 6 Not checked · 47 Own results and interpretations

32 more claims

Create a free account to see all 66 claims and to verify the 12 references still open.

See all claims — free account
  • Probably not a substantial claim. 17
  • It may state two separate claims as one. 14
  • Borderline: one of the tests for a substantial claim is unclear. 1

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

2 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

1 of 1

Citations

Cites

Show all 50 references

Research works printed but not linked: 5

  1. Songwei Ge, Thomas Hayes, Harry Yang, Xi Yin, Guan Pang, David Jacobs, Jia-Bin Huang, and Devi Parikh. 2022. Long Video Generation with Time-Agnostic VQGAN and Time-Sensitive Transformer. In Proceedings of the European Conference on Computer Vision (ECCV) (Lecture Notes in Computer Science), Shai Avidan, Gabriel Brostow, Moustapha Cissé, Giovanni Maria Farinella, and Tal Hassner (Eds.). Springer Nature Switzerland, Cham, 102--118.
  2. 10.5555/3327345.3327360 · doi:10.5555/3327345.3327360
  3. Yuezun Li and Siwei Lyu. 2019. Exposing DeepFake Videos By Detecting Face Warping Artifacts. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. 7.
  4. 10.5555/3618408.3619590 · doi:10.5555/3618408.3619590
  5. Yang Yu, Xiaolong Liu, Rongrong Ni, Siyuan Yang, Yao Zhao, and Alex C. Kot. 2023. PVASS-MDD: Predictive Visual-audio Alignment Self-supervision for Multimodal Deepfake Detection. IEEE Transactions on Circuits and Systems for Video Technology (2023), 1--1.

Cited by

See all 59

Metadata

Authors
Cai, Z., Ghosh, S., Adatia, A. P., Hayat, M., Dhall, A., Gedeon, T., & Stefanov, K.
Affiliations
  • Monash University (AU)
  • Curtin University (AU)
  • Indian Institute of Technology Ropar (IN)
  • Qualcomm (United States) (US)
  • Flinders University (AU)
Venue
Proceedings of the 32nd ACM International Conference on Multimedia
Publisher
ACM
Published
2024-10-28
Pages
7414-7423
OpenAlex
W4403791323
Cited by
59
References
50
Citations by year
2026: 28 · 2025: 27 · 2024: 4
FWCI
12.22
Type
conference paper
Language
en
Open access
Locations
  • ACM International Conference on Multimedia (ACM MM) (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • Monash University Research Portal (Monash University) (submittedVersion, cc-by)
  • arXiv (Cornell University) (cc-by)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
  • Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
Keywords
deepfake detection, video manipulation, sound manipulation, synthetic dataset, large-scale datasets, data generation pipeline

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: AV-Deepfake1M: A Large-Scale LLM-Driven Audio-Visual Deepfake Dataset. https://app.cochat.ai/papers/10.1145/3664647.3680795
The paper
Cai, Z., Ghosh, S., Adatia, A. P., Hayat, M., Dhall, A., Gedeon, T., & Stefanov, K. (2024). AV-Deepfake1M: A Large-Scale LLM-Driven Audio-Visual Deepfake Dataset. Proceedings of the 32nd ACM International Conference on Multimedia. https://doi.org/10.1145/3664647.3680795

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.