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

Authors: Jielun Peng, Yabin Wang, Yaqi Li, Jincheng Liu, Xiaopeng Hong, Athanasios V. Vasilakos

Published: 2026-07-28 10:23:36+00:00

Comment: First place in the General AIGC Audio-Video Detection Challenge at the IJCAI-ECAI 2026 DDL 2.0 Workshop

AI Summary

This paper introduces DAV-Det, a decoupled audio-visual AIGC detection system that addresses the limitations of existing methods that rely on audio-visual content correspondence, which the authors show does not consistently hold in general scenarios. DAV-Det independently models forensic evidence from each modality, utilizing decision-level fusion, and achieved first place in the General AIGC Audio-Video Detection Challenge.

Abstract

Generative AI has rapidly expanded audio-visual forgery beyond human-centric deepfakes into general scenes. Existing AIGC detection methods assume audio-visual content correspondence, identifying forgeries by spotting cross-modal inconsistencies. However, we empirically find that this assumption does not consistently hold in general scenarios. We argue that, for general audio-visual AIGC detection, decision-level fusion is a more robust alternative to feature-level fusion. Therefore, we propose DAV-Det, a decoupled audio-visual AIGC detection system that independently models forensic evidence from each modality. The visual detector leverages multi-granularity representations at global, patch, and segment levels to capture spatial forgery cues, while the audio detector exploits both temporal and spectral irregularities via a gated temporal-spectral dual-branch architecture to model acoustic artifacts. Our method ranks 1st in the General AIGC Audio-Video Detection Challenge of the IJCAI-ECAI 2026 DDL 2.0 Workshop, with a final score of 0.8460. Code is available at https://github.com/tuffy-studio/DAV-Det.

Key findings

The assumption of audio-visual content correspondence, central to many existing multi-modal deepfake detectors, does not consistently hold in general AIGC scenarios. DAV-Det, with its modality-decoupling and decision-level fusion, outperforms other methods in general AIGC audio-video detection, achieving a final score of 0.8460 and ranking 1st in the IJCAI-ECAI 2026 DDL 2.0 Workshop Challenge.

Approach

DAV-Det employs a decoupled architecture with independent visual and audio detectors. The visual detector captures spatial forgery cues using multi-granularity representations (global, patch, segment levels), while the audio detector models acoustic artifacts by exploiting temporal and spectral irregularities through a gated temporal-spectral dual-branch architecture. Predictions from both modalities are fused at the decision level.

Datasets

MVAD (General AIGC Audio–Video Detection benchmark DDL-GAV), FakeAVCeleb

Model(s)

DINOv3 ViT-L/16 (visual backbone), PEAV-base audio encoder (audio backbone)

Author countries

China, Norway

← Previous