MLN-WavLM：只用真实语音训练的开集音频深伪检测方法
When Normality Shifts: Multi-Level WavLM Modeling for Open-World Audio Deepfake Detection
做语音安全或反深伪的工程团队值得看这篇。它用 WavLM 加 Mahalanobis 距离，不碰攻击样本就能拿到 8.45% EER，和监督基线 7.86% 很接近。更关键是它诚实展示了跨域 22.12% EER 和阈值自适应失效的失败模式，对部署选型很有参考价值。
新加坡南洋理工等团队在 arXiv 发布论文（编号 2611.11465），提出 MLN-WavLM，一种攻击无关的音频深伪检测方法。该方法只在真实语音上训练，基于冻结的 WavLM Base+ 抽取的池化隐层状态做 shrinkage-Mahalanobis 建模，通过偏离程度给合成音频打分，避免检测器学到特定生成器的捷径。
在 ASVspoof 2019 LA 上，最优攻击无关变体取得 8.45% EER，接近全监督 WavLM-HM-Conformer 基线的 7.86%。但换到 ASVspoof 2021 DF 域偏移场景时，EER 升至 22.12%，说明真实语音校准的决策点在跨域时并不稳固。逐层分析显示 WavLM 早期层表征主导了集成效果。
团队还用无标签阈值自适应模拟真实部署流：域内污染不超过 10% 时，中位数/MAD 重校准能应对；但在 2021 DF 域偏移下重校准会导致低误报率召回全面崩塌。论文结论是冻结源域的正态性模型能排序深伪音频，但无法给出稳定的开集世界决策规则，下游阈值调整救不回上游排序的退化。
When Normality Shifts: Multi-Level WavLM Modeling for Open-World Audio Deepfake Detection
Audio deepfake detectors trained with known synthesis attacks can learn generator-specific shortcuts that fail as synthesis systems and acoustic conditions evolve. We investigate an attack-free alternative: a normality model trained exclusively on bona fide speech, which scores synthetic audio by its deviation from authentic representations. We propose MLN-WavLM, a method that applies shrinkage-Mahalanobis modeling to pooled hidden states from a frozen large-scale, pre-trained speech processing WavLM Base+ architecture. Evaluated on ASVspoof 2019 LA, our best attack-free variant achieves an 8.45\% Equal Error Rate (EER), closely tracking a fully supervised WavLM-HM-Conformer baseline (7.86%). However, when subjected to the ASVspoof 2021 DF domain shift, performance drops to 22.12% EER, exposing vulnerabilities in the bona-fide-calibrated operating point. Layer-wise analysis reveals that early WavLM representations dominate the ensemble. To test real-world viability, we simulated deployment streams using unlabeled threshold adaptation. While robust median/MAD recalibration easily handles in-domain contamination up to 10%, recalibrating under the 2021 DF domain shift triggers a total collapse of low-FPR recall. We conclude that while frozen source-domain normality can effectively rank deepfakes, it does not yield a stable open-world decision rule, and downstream threshold adaptation cannot rescue upstream ranking degradation.
This report is categorized under AITOP's "Papers" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include WavLM, 音频深伪检测, ASVspoof, 语音安全.
Source attribution: curated from arXiv cs.LG (@Daniel Niekurzak, Daniel Kozuch, Weronika Folwarska, Khan Muhammad, Rafał Scherer).