ScrambleMark evades speech watermarking via harmonic scrambling
Original title: ScrambleMark: Zero-Query Watermark-Agnostic Black-Box Evasion of Modern Speech Watermarking
ScrambleMark achieves 97% evasion on commercial platforms while preserving speech naturalness at 4.2/5.
ScrambleMark disrupts speech watermark consistency by perturbing harmonic structure rather than removing watermarks. The zero-query black-box attack combines a deterministic perturbation engine with a generative model for randomized scrambling patterns. It evades Google SynthID, Resemble AI, and TrustedAudio with nearly 97% success and high naturalness ratings. The method withstands adaptive recovery attempts designed to reverse it.
- Harmonic structure disruption breaks watermark consistency without quality loss
- ScrambleMark operates without watermark-specific rules or paired speech data
- Evasion exceeds 97% on three commercial platforms versus 1.1-1.7 for prior attacks
- Attack remains effective against adaptive recovery designed to reverse it
From the original article
1 min read at arxiv.orgSpeech watermarking is becoming a key defense for digital copyright and content authentication. Existing attacks treat evasion as a removal problem: they erase the watermark, relying on detector access or on speech data that is hard to collect, such as paired clean-watermarked or clean target speaker speech. Commercial platforms are moving to close off these assumptions.
By Binhao Ma · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.