HierCon: Hierarchical Contrastive Attention for Audio Deepfake Detection

TL;DR

HierCon introduces a hierarchical attention framework with contrastive learning to improve detection of audio deepfakes by modeling dependencies across layers and time, achieving state-of-the-art results and better cross-domain generalization.

Contribution

The paper presents HierCon, a novel hierarchical layer attention model with contrastive learning that captures dependencies across layers and time for more effective audio deepfake detection.

Findings

Achieves state-of-the-art EER of 1.93% on ASVspoof 2021 DF dataset.

Improves detection robustness across different domains and recording conditions.

Hierarchical modeling enhances generalization over independent layer approaches.

Abstract

Audio deepfakes generated by modern TTS and voice conversion systems are increasingly difficult to distinguish from real speech, raising serious risks for security and online trust. While state-of-the-art self-supervised models provide rich multi-layer representations, existing detectors treat layers independently and overlook temporal and hierarchical dependencies critical for identifying synthetic artefacts. We propose HierCon, a hierarchical layer attention framework combined with margin-based contrastive learning that models dependencies across temporal frames, neighbouring layers, and layer groups, while encouraging domain-invariant embeddings. Evaluated on ASVspoof 2021 DF and In-the-Wild datasets, our method achieves state-of-the-art performance (1.93% and 6.87% EER), improving over independent layer weighting by 36.6% and 22.5% respectively. The results and attention visualisations confirm that hierarchical modelling enhances generalisation to cross-domain generation techniques and recording conditions.