Audio-Anchored Fusion of Multi-Ratio DiT Reconstruction Residuals for Cross-Domain Audio Deepfake Detection

2026-07-29

AI Summary

This work addresses the significant degradation in generalization of existing audio deepfake detection models when faced with shifts in generators, corpora, or recording conditions. To enhance cross-domain robustness, the authors propose a reconstruction probing method based on a frozen Diffusion Transformer (DiT). Trained exclusively on genuine speech, the DiT generates reconstruction residuals under multi-rate masking. These residuals are then integrated into WavLM representations via a novel audio-anchored fusion mechanism, which applies them as scalar-gated additive corrections in a non-competitive manner, thereby avoiding gating-induced attenuation. The approach achieves EERs of 6.54% and 13.84% on ASVspoof 2021 Eval and In-the-Wild Full partitions, respectively, outperforming a specialized WavLM-ResNet18 baseline across three subsets on average and substantially improving cross-domain detection robustness.

Abstract

Audio deepfake detectors often degrade when generators, corpora, or recording conditions change. We use a Diffusion Transformer (DiT), trained only on bona fide speech, as a frozen reconstruction probe. Reconstructions at masking ratios 0.5, 0.75, and 0.9 yield explicit multi-ratio residual maps. Because these residuals are domain sensitive, our audio-anchored detector passes the projected frozen-WavLM auditory representation into the fusion sum without gate-based attenuation and uses residuals only as a scalar-gated additive correction. The pre-specified seed-42 run obtains 6.5442% EER / 0.18456 min-DCF on ASVspoof 5 Eval and 13.8372% / 0.36921 on ITW Full; three-seed means are 6.8885 (0.3308)% and 15.3328 (2.0719)%. The latter is below a separately optimized WavLM-ResNet18 reference under both supervision settings. Auxiliary supervision raises dynamic competitive fusion from 18.4007% to 25.2968% mean ITW EER, worsening all three seeds. The results support reconstruction residuals as complementary evidence and motivate a non-competitive auditory path for ASVspoof 5-to-ITW transfer, without claiming a componentwise causal ablation of anchoring alone.

Problem

• audio deepfake detection
• cross-domain
• domain shift
• reconstruction residuals
• robustness

Innovation

• Diffusion Transformer
• reconstruction residuals
• audio-anchored fusion
• cross-domain deepfake detection
• frozen representation