WST-X applies cascaded wavelet convolutions and modulus nonlinearities to produce stable multi-scale features that outperform prior front-ends on Deepfake-Eval-2024 and cross-dataset tests.
Bootstrap methods for standard errors, confidence intervals, and other measures of statistical accuracy.Statistical science, pages 54–75, 1986
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
WST-X Series: Wavelet Scattering Transform for Interpretable Speech Deepfake Detection
WST-X applies cascaded wavelet convolutions and modulus nonlinearities to produce stable multi-scale features that outperform prior front-ends on Deepfake-Eval-2024 and cross-dataset tests.