Audio deepfake detectors improve accuracy using controlled recurrent refinement
CoRELoop: Parameter-Efficient Controlled Recurrent Refinement for Audio Deepfake Detection
SoundArtificial Intelligence
Summary
Detecting fake audio is hard, especially when new kinds of fake sounds appear. The authors found a way to improve an existing audio detector by letting it refine its guess multiple times without changing its main parts or needing more training data. They created a method called CoReLoop that carefully controls this refinement process so it gets better with each pass. This helps the detector catch fake audio more accurately while keeping the original detector intact.
What this means in practice
- •For security teams: Deploy enhanced audio deepfake detectors that improve detection accuracy on unseen attacks without retraining the full model.
- •For audio software developers: Integrate lightweight refinement modules into audio authentication tools to maintain up-to-date detection with limited parameter changes.$Commercial implications: Enables sales of more reliable audio verification software by improving detection performance cost-effectively using existing models.
Authors
Kunyu Feng, Yuxiang Wang, Li Wang, Wan Lin, Zhizheng Wu
Abstract
Generalizing to unseen attacks remains challenging for audio deepfake detectors, and collecting training data covering all potential attacks is impractical. We explore recurrent refinement in an already-trained SSL-based detector without additional data or changes to its original parameters. However, directly recycling encoder outputs as inputs degrades detection in our diagnostic. We propose CoReLoop, which makes this reuse effective by adapting recurrent inputs to the frozen encoder, controlling state updates, and aligning refined outputs with the frozen classifier. By training only lightweight refinement modules and loop-specific low-rank adapters on the original data, CoReLoop enables additional refinement while preserving the detector's original first-pass prediction. On 14 cross-domain test sets, the 24-layer model reduces pooled equal error rate (EER) from 4.85% to 3.74% with two passes, with approximately 10M trainable parameters out of 598M. To selectively apply this refinement, an optional halting head chooses the depth for each utterance, achieving 3.73% pooled EER with an average of 1.18 passes.