Generalizable speech deepfake detection via meta-learned LoRA

TL;DR

This paper introduces a meta-learning approach using LoRA adapters for speech deepfake detection, achieving strong zero-shot performance and robustness across different attack types with minimal parameter updates.

Contribution

It presents a novel combination of LoRA adapters and meta-learning for domain generalization in speech deepfake detection, significantly reducing training parameters while improving cross-domain accuracy.

Findings

Achieves lower EER (5.30%) compared to full fine-tuning (8.84%).

Requires only 1.1% of parameters compared to full fine-tuning.

Outperforms fully fine-tuned models on five of six evaluation corpora.

Abstract

Reliable detection of speech deepfakes (spoofs) must remain effective when the distribution of spoofing attacks shifts. We frame the task as domain generalization and show that inserting Low-Rank Adaptation (LoRA) adapters into every attention head of a self-supervised (SSL) backbone, then training only those adapters with Meta-Learning Domain Generalization (MLDG), yields strong zero-shot performance. The resulting model updates about 3.6 million parameters, roughly 1.1% of the 318 million updated in full fine-tuning, yet surpasses a fully fine-tuned counterpart on five of six evaluation corpora. A first-order MLDG loop encourages the adapters to focus on cues that persist across attack types, lowering the average EER from 8.84% for the fully fine-tuned model to 5.30% with our best MLDG-LoRA configuration. Our findings show that combining meta-learning with parameter-efficient adaptation offers an effective method for zero-shot, distribution-shift-aware speech deepfake detection.