defense arXiv Sep 28, 2025 · Sep 2025
Pu Huang, Shouguang Wang, Siya Yao et al. · Zhejiang Gongshang University · New Jersey Institute of Technology
Novel speech deepfake detector combining information bottleneck and confidence-aware adversarial alignment for generalizable detection across unseen spoofing methods
Output Integrity Attack audio
Neural speech synthesis techniques have enabled highly realistic speech deepfakes, posing major security risks. Speech deepfake detection is challenging due to distribution shifts across spoofing methods and variability in speakers, channels, and recording conditions. We explore learning shared discriminative features as a path to robust detection and propose Information Bottleneck enhanced Confidence-Aware Adversarial Network (IB-CAAN). Confidence-guided adversarial alignment adaptively suppresses attack-specific artifacts without erasing discriminative cues, while the information bottleneck removes nuisance variability to preserve transferable features. Experiments on ASVspoof 2019/2021, ASVspoof 5, and In-the-Wild demonstrate that IB-CAAN consistently outperforms baseline and achieves state-of-the-art performance on many benchmarks.
transformer Zhejiang Gongshang University · New Jersey Institute of Technology
ML Security Papers