Research map / Adversarial robustness and security / Attacks on text and graphs
Yours or Mine? Overwriting Attacks Against Neural Audio Watermarking
Lingfeng Yao, Chenpei Huang, Shengyao Wang, Junpei Xue, Hanqing Guo, Jiang Liu, Phone Lin, Tomoaki Ohtsuki, Miao Pan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| overwrite with the same watermarking method (white-box) | AudioSeal, Timbre, WavMark · white-box overwriting | BER of original watermark: BER near 0.5 (0.4931 / 0.4990 / 0.4981) | Improved |
| overwrite with a different watermarking method than the original (ablation) | overwriting attack · white-box cross-method | BER of original watermark: BER 0.0006 | Hurt |
| use a surrogate model trained with a different pipeline (cross-training) (ablation) | gray-box overwriting attack · VoxCeleb1, AudioSeal/Timbre/WavMark | ASR: ASR 98.40 to 100.00 | Improved |
| use a surrogate trained on a different dataset (cross-data) (ablation) | gray-box overwriting attack · LibriSpeech surrogate vs VoxCeleb1 original | ASR: ASR 98.80 to 100.00 | Improved |
| stack more embedders in zero-query attack (ablation) | black-box zero-query attack · black-box | ASR: nearly 30% -> almost 100% | Improved |
| stack more embedders in zero-query attack (ablation) | black-box zero-query attack · black-box | SNR (dB): about 24 dB -> 20 dB | Hurt |
| use query-based attack instead of zero-query | black-box overwriting attack · black-box | training iterations: 36,000 iters -> 14,000 iters | Improved |
| use query-based attack instead of zero-query | black-box overwriting attack · black-box | SNR (dB): 20.63 -> 24.19 | Improved |
Closest papers by meaning
- XAttnMark: Learning Robust Audio Watermarking with Cross-Attention (ICML 2025 · similarity 0.88)
- AudioMarkBench: Benchmarking Robustness of Audio Watermarking (NeurIPS 2024 · similarity 0.87)
- Asymmetric Phase Coding Audio Watermarking (NeurIPS 2026 · similarity 0.84)
- Watermark Stealing in Large Language Models (ICML 2024 · similarity 0.83)
- Robust Distortion-Free Watermark for Autoregressive Audio Generation Models (NeurIPS 2025 · similarity 0.83)
- Proactive Detection of Voice Cloning with Localized Watermarking (ICML 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
