Less is More: Modality-Decoupling for General AIGC Audio-Video Detection

Researchers proposed DAV-Det, a general audio-video AIGC detector that decouples modalities and analyzes audio and video separately.

Instead of depending on cross-modal consistency, the model learns independent visual and audio forensic cues, then fuses decisions at the end.

It uses multi-level visual representations and a dual-branch audio design to better detect forged or generated content.

The system ranked first in a benchmark challenge, highlighting the need for more robust detection beyond traditional deepfake methods.