ML Security PapersWST-X Series: Wavelet Scattering Transform for Interpretable Speech Deepfake Detection
Xi Xuan 1, Davide Carbone 2, Ruchi Pandey 1, Wenxin Zhang 3,4, Tomi Kinnunen 1
1 University of Eastern Finland
2 Laboratoire de Physique de l'Ecole Normale Supérieure
Published on arXiv
2602.02980
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
WST-X outperforms existing hand-crafted and SSL-based front-ends by a wide margin on the Deepfake-Eval-2024 benchmark, with small averaging scale J combined with high frequency and directional resolutions (Q, L) being critical for artifact capture.
WST-X
Novel technique introduced
Designing front-ends for speech deepfake detectors primarily focuses on two categories. Hand-crafted filterbank features are transparent but are limited in capturing high-level semantic details, often resulting in performance gaps compared to self-supervised (SSL) features. SSL features, in turn, lack interpretability and may overlook fine-grained spectral anomalies. We propose the WST-X series, a novel family of feature extractors that combines the best of both worlds via the wavelet scattering transform (WST), integrating wavelets with nonlinearities analogous to deep convolutional networks. We investigate 1D and 2D WSTs to extract acoustic details and higher-order structural anomalies, respectively. Experimental results on the recent and challenging Deepfake-Eval-2024 dataset indicate that WST-X outperforms existing front-ends by a wide margin. Our analysis reveals that a small averaging scale ($J$), combined with high-frequency and directional resolutions ($Q, L$), is critical for capturing subtle artifacts. This underscores the value of translation-invariant and deformation-stable features for robust and interpretable speech deepfake detection.
Key Contributions
- First application of the wavelet scattering transform (WST) as a front-end feature extractor for speech deepfake detection, requiring no training data
- Investigation of both 1D and 2D WST variants to capture acoustic artifacts and higher-order structural anomalies in synthetic speech
- Systematic analysis of WST hyperparameters (J, Q, L) and integration strategies (parallel and cascaded) with SSL features, demonstrating superiority over existing front-ends on Deepfake-Eval-2024
🛡️ Threat Analysis
Proposes a novel feature extractor family (WST-X) for detecting AI-generated/synthetic speech — this is AI-generated content detection (deepfake detection), a core ML09 output integrity concern. The contribution is a novel detection methodology, not merely applying existing methods to a new domain.