Research

Paper

TESTING March 06, 2026

Do Compact SSL Backbones Matter for Audio Deepfake Detection? A Controlled Study with RAPTOR

Authors

Ajinkya Kulkarni, Sandipana Dowerah, Atharva Kulkarni, Tanel Alumäe, Mathew Magimai Doss

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.

Metadata

arXiv ID: 2603.06164
Provider: ARXIV
Primary Category: cs.SD
Published: 2026-03-06
Fetched: 2026-03-09 06:05

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.06164v1</id>\n    <title>Do Compact SSL Backbones Matter for Audio Deepfake Detection? A Controlled Study with RAPTOR</title>\n    <updated>2026-03-06T11:16:55Z</updated>\n    <link href='https://arxiv.org/abs/2603.06164v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.06164v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>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.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.SD'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-03-06T11:16:55Z</published>\n    <arxiv:comment>Submitted to Interspeech 2026, 4 pages, 2 figures</arxiv:comment>\n    <arxiv:primary_category term='cs.SD'/>\n    <author>\n      <name>Ajinkya Kulkarni</name>\n    </author>\n    <author>\n      <name>Sandipana Dowerah</name>\n    </author>\n    <author>\n      <name>Atharva Kulkarni</name>\n    </author>\n    <author>\n      <name>Tanel Alumäe</name>\n    </author>\n    <author>\n      <name>Mathew Magimai Doss</name>\n    </author>\n  </entry>"
}