Communication Dans Un Congrès Année : 2025

KAN You Hear the Truth? Audio Deepfake Detection with Kolmogorov-Arnold Networks

Résumé

Kolmogorov-Arnold Networks (KAN) have recently emerged as a promising alternative to traditional multilayer perceptrons, offering superior performance and greater interpretability. In this work, we explore the potential of KAN by demonstrating its ability to enhance audio deepfake detection and robustness. To further improvement, we employ multi-level token classification by leveraging speech representations of foundation model. Experimental results across multiple evaluation datasets demonstrate that our approach enhances both specialization and generalization. These findings provide potential insights into the integration of KAN, laying the way for future research in this domain.

Fichier principal
Vignette du fichier
EUSIPCO_2025_camera_ready_version.pdf (2.8 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05283051 , version 1 (29-09-2025)

Licence

Identifiants

  • HAL Id : hal-05283051 , version 1

Citer

Hoan My Tran, Damien Lolive, David Guennec, Aghilas Sini, Arnaud Delhay, et al.. KAN You Hear the Truth? Audio Deepfake Detection with Kolmogorov-Arnold Networks. EUSIPCO 2025, EURASIP, Sep 2025, Palerme, Italy. pp.561-565. ⟨hal-05283051⟩
834 Consultations
242 Téléchargements

Partager

  • More