AWARE: Audio Watermarking with Adversarial Resistance to Edits

TL;DR

AWARE tackles robustness of audio watermarking to edits by coupling adversarial embedding in the time-frequency domain with a detector that is invariant to temporal misalignment. It introduces a per-bin perceptual budget and a Bitwise Readout Head to aggregate temporal evidence, enabling reliable bit decoding under desynchronization and cuts. Empirical results show strong perceptual quality and consistently low BER across diverse edits, often outperforming representative baselines without heavy attack-stack augmentation. This architecture-centered approach offers a principled path for robust audio provenance and content protection in realistic, edit-prone pipelines.

Abstract

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.

Figures (1)

• Figure 1: AWARE detector architecture.