Spoofed speech detection with a focus on speaker embedding - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Communication Dans Un Congrès Année : 2024

Spoofed speech detection with a focus on speaker embedding

Résumé

Self-Supervised Learning (SSL) models excel as feature ex-tractors in downstream speech tasks, including the increasingly crucial area of spoof speech detection due to the rise of audio deepfakes using Text-To-Speech (TTS) and Voice Conversion(VC) technologies. To address this issue, we propose a novel approach that relies on speaker embedding using a fine tuned WavLM model with layer-wise attentive statistics pooling combined to a supervised contrastive learning and cross-entropy loss. Evaluation on Logical Access (LA) and DeepFake (DF)tasks on ASVspoof 2019 and 2021 highlights its potential in detecting audio deepfakes, with the contrastive loss producing more stable results among test sets
Fichier non déposé

Dates et versions

hal-04651084 , version 1 (17-07-2024)

Identifiants

  • HAL Id : hal-04651084 , version 1

Citer

Hoan My Tran, David Guennec, Philippe Martin, Aghilas Sini, Damien Lolive, et al.. Spoofed speech detection with a focus on speaker embedding. INTERSPEECH 2024, Sep 2024, Kos Island, Greece. ⟨hal-04651084⟩
0 Consultations
0 Téléchargements

Partager

Gmail Mastodon Facebook X LinkedIn More