Averaging two audio copies breaks traceability in personalized watermarking systems
Original title: Averaging Collusion Breaks Traceability in Personalized Neural Audio Watermarking
Averaging only two personalized audio copies produces tracing-failure rates of 87.7-99.7% while largely preserving audio quality.
Personalized neural audio watermarking systems fail when multiple copies of the same recording become available. Simple sample-wise averaging of just two copies breaks traceability across five systems. The attack preserves audio quality and can be made more sophisticated with payload-aware mixture weighting.
- Tracing failure rates range from 87.7% to 99.7% for two-copy averaging
- Decoder outputs show distinct responses to coalition payload composition
- Mixture weighting steers the decoder toward a selected nonmember at 92.3% success in TimbreWM
- Code and results are available on GitHub
Impact: Breaking traceability of audio watermarking by averaging multiple copies
From the original article
1 min read at arxiv.orgPersonalized neural audio watermarking systems embed user-specific payloads into audio to trace the user associated with a leaked or misused copy. Existing evaluations mainly assume access to a single personalized copy, but real-world distribution may expose multiple copies of the same recording.
By Yangming Liu · 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.