New discrete optimal transport attack targets speaker verification systems

Researchers have developed a novel adversarial attack method using discrete optimal transport (DOT) that targets automatic speaker verification (ASV) and anti-spoofing systems. This black-box attack operates by aligning frame-level embeddings of generated or other speech towards bona fide speech regions in the representation space, rather than directly maximizing speaker similarity. Experiments on ASVspoof2019 and ASVspoof5 datasets show that the DOT attack significantly degrades ASV performance and increases countermeasure error rates, demonstrating its effectiveness across different datasets and even after countermeasure fine-tuning. AI

排序理由 Research paper detailing a new adversarial attack method. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

• Anton Selitskiy
• ASVspoof2019
• ASVspoof5
• Discrete optimal transport
• Fréchet Audio Distance
• WavLM

• 论文
• 安全

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →