Reproduce “IDEAW: Robust Neural Audio Watermarking with Invertible Dual-Embedding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pengcheng Li, Xulong Zhang, Jing Xiao, Jianzong Wang
The IDEAW model presents a neural audio watermarking technique utilizing an invertible dual-embedding strategy to improve watermark localization efficiency and a balance block to maintain INN symmetry during adversarial training. It achieves high imperceptibility and robustness, demonstrating up to 50% faster watermark locating compared to previous methods.
Key insights
- Separating the embedding of a lightweight locating code from the main message significantly improves watermark localization efficiency by enabling a rapid pre-check.
- A "balance block" can effectively address the challenge of maintaining Invertible Neural Network (INN) symmetry and stability when an attack layer is present, enhancing robustness.
- Adversarial training with a discriminator is crucial for ensuring the perceptual indistinguishability of watermarked audio from original audio.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.