arXiv cs.CL· Ajinkya Kulkarni, Sandipana Dowerah, Atharva Kulkarni, Tanel Alum\"ae, Mathew Magimai Doss·· 14 小时前AI 评分38
RAPTOR 受控研究：紧凑 SSL 骨干是否足以支撑音频深度伪造检测
Do Compact SSL Backbones Matter for Audio Deepfake Detection? A Controlled Study with RAPTOR
AI 导读
研究提出 RAPTOR——基于成对门控融合的紧凑 SSL 骨干受控研究，在 14 个跨域基准上评测 HuBERT 与 WavLM 变体。结果显示多语言 HuBERT 预训练是跨域鲁棒性的主要来源，使 100M 参数模型可匹敌更大规模及商用系统。研究还引入带扰动不确定性估计的测试时增强协议，发现 WavLM 变体在扰动下过度自信、校准失准，而迭代 mHuBERT 保持稳定。
正文 · 原文
Abstract:Self-supervised learning (SSL) underpins modern audio deepfake detection, yet most prior work centers on a single large wav2vec2-XLSR backbone, leaving compact under studied. We present RAPTOR, Representation Aware Pairwise-gated Transformer for Out-of-domain Recognition a controlled study of compact SSL backbones from the HuBERT and WavLM within a unified pairwise-gated fusion detector, evaluated across 14 cross-domain benchmarks. We show that multilingual HuBERT pre-training is the primary driver of cross-domain robustness, enabling 100M models to match larger and commercial systems. Beyond EER, we introduce a test-time augmentation protocol with perturbation-based aleatoric uncertainty to expose calibration differences invisible to standard metrics: WavLM variants exhibit overconfident miscalibration under perturbation, whereas iterative mHuBERT remains stable. These findings indicate that SSL pre-training trajectory, not model scale, drives reliable audio deepfake detection.
| Subjects: | Sound (cs.SD); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2603.06164 [cs.SD] |
| (or arXiv:2603.06164v2 [cs.SD] for this version) | |
| https://doi.org/10.48550/arXiv.2603.06164 arXiv-issued DOI via DataCite |
Submission history
From: Ajinkya Kulkarni [view email]
[v1]
Fri, 6 Mar 2026 11:16:55 UTC (315 KB)
[v2]
Sun, 25 Oct 2026 14:58:36 UTC (315 KB)
来源：arXiv cs.CL · arxiv.org