defense arXiv Sep 17, 2025 · Sep 2025
Janne Laakkonen, Ivan Kukanov, Ville Hautamäki · University of Eastern Finland · KLASS Engineering and Solutions
Novel Mixture-of-LoRA-Experts architecture improves generalization of audio deepfake detectors to unseen synthesis attacks
Output Integrity Attack audio
Foundation models such as Wav2Vec2 excel at representation learning in speech tasks, including audio deepfake detection. However, after being fine-tuned on a fixed set of bonafide and spoofed audio clips, they often fail to generalize to novel deepfake methods not represented in training. To address this, we propose a mixture-of-LoRA-experts approach that integrates multiple low-rank adapters (LoRA) into the model's attention layers. A routing mechanism selectively activates specialized experts, enhancing adaptability to evolving deepfake attacks. Experimental results show that our method outperforms standard fine-tuning in both in-domain and out-of-domain scenarios, reducing equal error rates relative to baseline models. Notably, our best MoE-LoRA model lowers the average out-of-domain EER from 8.55\% to 6.08\%, demonstrating its effectiveness in achieving generalizable audio deepfake detection.
transformer gnn University of Eastern Finland · KLASS Engineering and Solutions
ML Security Papers