ISCA Archive Interspeech 2024
ISCA Archive Interspeech 2024

DGPN: A Dual Graph Prototypical Network for Few-Shot Speech Spoofing Algorithm Recognition

Zirui Ge, Xinzhou Xu, Haiyan Guo, Tingting Wang, Zhen Yang, Björn W. Schuller

As synthetic speech technologies rapidly advance, accurately classifying these synthesis algorithms has become increasingly critical in the speech anti-spoofing. Nevertheless, in the incipient stage of emerging spoofing algorithms, the acquisition of ample generated speech samples is often constrained, impeding the efficacy of conventional models. To this end, we introduce a novel methodology within the realm of few-shot learning, named Dual Graph Prototypical Network (DGPN), in view of this limitation for the Speech Spoofing Algorithm Recognition (SSAR) task. The proposed method consists of intra-speech graph and inter-speech graph modules, where the former employs graph attention networks to model the low-level representations of an utterance, and the latter utilizes graph neural networks to depict high-level representations of different utterances. Experimental evaluations demonstrate that the proposed method outperforms existing models in classification accuracy, showcasing its effectiveness in addressing the challenge of the few-shot SSAR task.