Effects of Accessible Renderings on Deepfake Detection

Determine whether captions, subtitles, or sign-language interpretation provide useful contextual evidence, remove diagnostic cues, or introduce artifacts that viewers mistakenly interpret as evidence of manipulation when evaluating audiovisual deepfakes.

Background

The study evaluated participants using the same unmediated audiovisual signal and did not include captions, subtitles, or sign-language interpretation during the detection task. Consequently, the findings do not establish how accessibility-mediated representations affect the evidence available to DHH viewers when judging authenticity.

This unresolved problem is important because captions and interpretation may simultaneously improve access to manipulated messages, omit acoustic or other diagnostic information, and introduce rendering errors. Determining their effects would support the design of accessible deepfake-detection and verification systems for DHH users.

References

We therefore cannot determine whether an accessible rendering adds useful context, removes diagnostic cues, or introduces artifacts mistaken for manipulation.

— "What I See is What I Hear": Deepfake Detection Across Diverse Hearing Abilities  (2609.28659 - Pasternak et al., 23 Sep 2026) in Section VII-G, Limitations