The authors recorded 496 consenting participants speaking English, German, Italian, French or Spanish through personal devices in home environments.
Multilingual benchmark exposes deepfake detectors’ reliance on synthetic audio cues
BabelFake combines modern video and voice manipulations across five languages, revealing that detector performance depends strongly on how audio and video are generated.
Industry
Carlotta Segna · Joel Tschesche · Anna Rohrbach
TU Darmstadt · Hessian.AI
Research Digest··2 min read
Segna, Tschesche and Rohrbach introduce a consent-based benchmark for testing audio-visual DeepFake detectors across languages, manipulation methods and demographic groups.
Why this paper
From TU Darmstadt and Hessian.AI
In one line
BabelFake shows that audio-visual DeepFake detectors degrade when manipulated video retains authentic audio, with robustness also varying by language, demographics, architecture, and training data.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§