benchmark arXiv Mar 6, 2026 · 4w ago
Ajinkya Kulkarni, Sandipana Dowerah, Atharva Kulkarni et al. · Idiap Research Institute · Tallinn University of Technology +1 more
Controlled study benchmarking compact SSL backbones for audio deepfake detection with TTA-based uncertainty calibration
Output Integrity Attack audio
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.
transformer Idiap Research Institute · Tallinn University of Technology · MBZUAI
ML Security Papers