Saved in:
Bibliographic Details
Main Authors: Teissier, Antoine, Tahon, Marie, Dugué, Nicolas, Sini, Aghilas
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.05696
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915535858434048
author Teissier, Antoine
Tahon, Marie
Dugué, Nicolas
Sini, Aghilas
author_facet Teissier, Antoine
Tahon, Marie
Dugué, Nicolas
Sini, Aghilas
contents Due to the rapid progress of speech synthesis, deepfake detection has become a major concern in the speech processing community. Because it is a critical task, systems must not only be efficient and robust, but also provide interpretable explanations. Among the different approaches for explainability, we focus on the interpretation of latent representations. In such paper, we focus on the last layer of embeddings of AASIST, a deepfake detection architecture. We use a TopK activation inspired by SAEs on this layer to obtain sparse representations which are used in the decision process. We demonstrate that sparse deepfake detection can improve detection performance, with an EER of 23.36% on ASVSpoof5 test set, with 95% of sparsity. We then show that these representations provide better disentanglement, using completeness and modularity metrics based on mutual information. Notably, some attacks are directly encoded in the latent space.
format Preprint
id arxiv_https___arxiv_org_abs_2510_05696
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Sparse deepfake detection promotes better disentanglement
Teissier, Antoine
Tahon, Marie
Dugué, Nicolas
Sini, Aghilas
Sound
Artificial Intelligence
Machine Learning
Due to the rapid progress of speech synthesis, deepfake detection has become a major concern in the speech processing community. Because it is a critical task, systems must not only be efficient and robust, but also provide interpretable explanations. Among the different approaches for explainability, we focus on the interpretation of latent representations. In such paper, we focus on the last layer of embeddings of AASIST, a deepfake detection architecture. We use a TopK activation inspired by SAEs on this layer to obtain sparse representations which are used in the decision process. We demonstrate that sparse deepfake detection can improve detection performance, with an EER of 23.36% on ASVSpoof5 test set, with 95% of sparsity. We then show that these representations provide better disentanglement, using completeness and modularity metrics based on mutual information. Notably, some attacks are directly encoded in the latent space.
title Sparse deepfake detection promotes better disentanglement
topic Sound
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2510.05696