Preprint
Jul 2026
Positive-Incentive Noise Predictor for Adversarial Purification in Speaker Verification
The Positive-Incentive Noise Predictor (PnP) is proposed, the first framework that explicitly introduces positive-incentive noise ({\pi}-noise) into the purification task and effectively defends against adversarial attacks while preserving performance on natural speech.
Yibo Bai, Sizhou Chen, M. Panariello et al.
· 0 citations