analiza aiqbench
This research tackles the critical generalization problem in audio deepfake detection by introducing a dual-granularity orthogonal disentanglement framework that separates speaker identity from synthesis artifacts without architectural complexity. By enforcing feature independence at both sample and batch levels, the method significantly lowers equal error rates across multiple datasets compared to existing adversarial approaches. The study represents a substantive advancement in machine learning robustness for audio forensics rather than an incidental mention of AI concepts.
Dual-Granularity Orthogonal Disentanglement for Generalizable Audio Deepfake Detection
arXiv:2606.16532v1 Announce Type: cross Abstract: Audio deepfake detectors often fail to generalize across speakers, as they learn speaker-identity features rather than synthesis artifacts, known as implicit identity leakage. Existing methods address this but incur architectural complexity or training instability. This paper proposes a dual-granularity orthogonal disentanglement framework enforcing feature independence at two levels: sample-level cosine orthogonality captures directional decorrelation, while batch-level cross-covariance regularization eliminates linear correlations across embedding dimensions. A curriculum disentanglement schedule progressively strengthens the orthogonality constraint without auxiliary networks or adversarial dynamics. Experiments on ASVspoof 2019 LA, ASVspoof 2021 DF, and In-the-Wild datasets demonstrate that the proposed method achieves 1.35%, 7.88%, and 21.58% equal error rates (EER), respectively, surpassing gradient reversal disentanglement by 2.60% absolute on cross-dataset transfer.
Abstract:Audio deepfake detectors often fail to generalize across speakers, as they learn speaker-identity features rather than synthesis artifacts, known as implicit identity leakage. Existing methods address this but incur architectural complexity or training instability. This paper proposes a dual-granularity orthogonal disentanglement framework enforcing feature independence at two levels: sample-level cosine orthogonality captures directional decorrelation, while batch-level cross-covariance regularization eliminates linear correlations across embedding dimensions. A curriculum disentanglement schedule progressively strengthens the orthogonality constraint without auxiliary networks or adversarial dynamics. Experiments on ASVspoof 2019 LA, ASVspoof 2021 DF, and In-the-Wild datasets demonstrate that the proposed method achieves 1.35%, 7.88%, and 21.58% equal error rates (EER), respectively, surpassing gradient reversal disentanglement by 2.60% absolute on cross-dataset transfer.
Sound (cs.SD); Artificial Intelligence (cs.AI)
arXiv:2606.16532 [cs.SD]
(or arXiv:2606.16532v1 [cs.SD] for this version)
https://doi.org/10.48550/arXiv.2606.16532 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 211 cuvinte
Subiecte conexe
Less is More: Modality-Decoupling for General AIGC Audio-Video Detection
arXiv · 29 iul. 2026
A Multi-Domain Feature Fusion Framework for Generalizable Deepfake Detection Across Different Generators
arXiv · 15 iun. 2026
SONAR: Spectral-Contrastive Audio Residuals for Generalizable Deepfake Detection
arXiv · 21 iul. 2026
Robust Workflow Generation via Adversarial Learning for Audio Deepfake Detection
arXiv · 18 sept. 2026