AWARE: Audio Watermarking with Adversarial Resistance to Edits
This addresses the need for reliable audio watermarking resistant to edits, offering a novel method that avoids overfitting to simulated distortions, though it is incremental in the field of learning-based watermarking.
The paper tackles the problem of robust audio watermarking by introducing AWARE, which uses adversarial optimization and a time-order-agnostic detector to achieve high audio quality and low bit error rates across various edits, often outperforming state-of-the-art systems.
Prevailing practice in learning-based audio watermarking is to pursue robustness by expanding the set of simulated distortions during training. However, such surrogates are narrow and prone to overfitting. This paper presents AWARE (Audio Watermarking with Adversarial Resistance to Edits), an alternative approach that avoids reliance on attack-simulation stacks and handcrafted differentiable distortions. Embedding is obtained via adversarial optimization in the time-frequency domain under a level-proportional perceptual budget. Detection employs a time-order-agnostic detector with a Bitwise Readout Head (BRH) that aggregates temporal evidence into one score per watermark bit, enabling reliable watermark decoding even under desynchronization and temporal cuts. Empirically, AWARE attains high audio quality and speech intelligibility (PESQ/STOI) and consistently low BER across various audio edits, often surpassing representative state-of-the-art learning-based audio watermarking systems.