CoRELoop: Parameter-Efficient Refinement for Audio Deepfake Detection
September 18, 2026
CoRELoop uses lightweight refinement modules and low-rank adapters to improve audio deepfake detection without retraining the base model. On 14 cross-domain test sets, the method reduced the equal error rate from 4.85% to 3.74% using only 10M trainable parameters.
HOW THIS AFFECTS YOU
●
builderYou can upgrade existing SSL-based detectors to handle unseen attacks with minimal parameter updates.
●
researcherThe approach demonstrates how recurrent refinement can be controlled to preserve original classifier performance.