Paper proposes robust feature-vocoder adversarial attacks on ASR

Arxiv CS.SD (Sound/Audio ML)

Attack adds perturbations to feature representations and reconstructs via vocoder, bypassing waveform-based defenses. Evaluated on multilingual ASR systems for robustness.

speech-to-textadversarial-attack

Arxiv CS.SD (Sound/Audio ML)·Yifan LiaoZongmin ZhangZhen SunYuhui SunXinhu ZhengXinlei He

6 days ago