Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

WST-X Series: Wavelet Scattering Transform for Interpretable Speech Deepfake Detection

Xi Xuan, Davide Carbone, Ruchi Pandey, Wenxin Zhang, Tomi H. Kinnunen

TL;DR

Speech deepfake detectors require front-ends that are both interpretable and robust. This paper proposes WST-X, a family of front-ends that fuse wavelet scattering transform representations with SSL latent features, explored in parallel (WST-X1) and cascaded (WST-X2) configurations. On the DE2024 benchmark, WST-X1 and WST-X2 outperform traditional DSP front-ends and PT-XLSR SSL baselines, with key insights showing that a small averaging scale $J$ together with high-frequency resolution $Q$ and directional resolution $L$ better reveal subtle artifacts. The work demonstrates translation-invariant, deformation-stable, interpretable features that improve detection and offer visual interpretability of artifacts, pointing toward applications in forensic traceability and future deepfake-source localization.

Abstract

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.

WST-X Series: Wavelet Scattering Transform for Interpretable Speech Deepfake Detection

TL;DR

Speech deepfake detectors require front-ends that are both interpretable and robust. This paper proposes WST-X, a family of front-ends that fuse wavelet scattering transform representations with SSL latent features, explored in parallel (WST-X1) and cascaded (WST-X2) configurations. On the DE2024 benchmark, WST-X1 and WST-X2 outperform traditional DSP front-ends and PT-XLSR SSL baselines, with key insights showing that a small averaging scale together with high-frequency resolution and directional resolution better reveal subtle artifacts. The work demonstrates translation-invariant, deformation-stable, interpretable features that improve detection and offer visual interpretability of artifacts, pointing toward applications in forensic traceability and future deepfake-source localization.

Abstract

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 (), combined with high-frequency and directional resolutions (), is critical for capturing subtle artifacts. This underscores the value of translation-invariant and deformation-stable features for robust and interpretable speech deepfake detection.
Paper Structure (14 sections, 2 equations, 3 figures, 2 tables)

This paper contains 14 sections, 2 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Wavelet Scattering Transform Theory
  4. Physical Interpretation of WST Parameters
  5. WST-X Series Feature Extractor
  6. Experimental setup
  7. Real-world Dataset Description and Evaluation Metrics
  8. Model Configurations
  9. Results and analysis
  10. Analysis of WST Parameters (RQ1)
  11. WST-X Series vs. Mel & Linear & CQ Filterbank (RQ2)
  12. WST-X Series vs. PT-XLSR (RQ3)
  13. Visualization and Interpretability
  14. Conclusions

Figures (3)

  • Figure 1: Hierarchical architecture of the second-order wavelet scattering transform, showing the extraction of zeroth-, first-, and second-order coefficients.
  • Figure 2: Overview of the WST-X Series: WST-X1 and WST-X2 feature extractors. The top panel illustrates Strategy I (parallel integration with 1D WST), while the bottom panel shows Strategy II (cascaded integration with 2D WST). GAP (Global Average Pooling); LP (Linear Projection); TE (Temporal Expansion); SF (Spatial Flattening).
  • Figure 3: Representations of a real utterance (top row) and a fake utterance synthesized by Qwen2.5-Omni (bottom row) across different front-ends: (a) Mel, (b) Linear, (c) Constant-Q Filterbank, (d) First-order WST, and (e) Second-order WST. The displayed WST representations correspond to the configuration $(J, Q) = (2, 10)$. Focusing on the bottom row, the correspondence between larger WST scales and lower spectrogram frequencies is visually evident. Notably, the preceding three spectrogram representations appear more coarse-grained than the first- and second-order WST, despite similar overall heatmap patterns. The blue bounding boxes highlight the visually distinctive parts of the fake speech signals within the WST features compared to real speech.