Neural audio watermarks stay compatible while reducing false alarms
Backward Compatibility in Zero-Bit Neural Audio Watermarking
SoundMachine LearningMultimedia
Summary
Audio watermarking hides secret signals in sounds to verify authenticity, but updates can break detection or cause false alarms. The authors designed a new training method to keep new detectors working well with older watermarks and avoid mixing them up with unrelated ones. They found pairing new detectors with frozen old embedders causes confusion and high false positives. To fix this, they introduced a technique that forces new watermarks to be distinct from old ones, improving accuracy without hurting audio quality.
What this means in practice
- •For digital content distributors: Maintain reliable detection of audio watermarks across multiple software versions without increased false alarms for each update.
- •For multimedia security teams: Securely verify watermarked audio even when embedding and detecting systems are selectively updated or decoupled to enhance security.
Authors
Admir Krnić, Kosta Pavlović, Petar Nedić, Željko Bolević, Slavko Kovačević, Igor Djurović
Abstract
Neural watermarking systems are typically trained as tightly coupled embedder-detector pairs and implicitly assume static deployment. In real-world deployments, embedders and detectors are frequently updated and selectively decoupled for security and versioning purposes, which can break detection in previously embedded content and, more critically, increase false-positive rates on embedders that a given detector is intentionally not intended to recognize. In this paper the concept of backward compatibility in zero-bit audio watermarking is formalized. A training framework is proposed that pairs a new detector with a sequence of frozen legacy embedders (its legacy chain), using a performance-weighted sampling strategy so that the detector retains accuracy across all earlier versions. This coupling, while necessary for backward compatibility, is shown to induce an embedder-collapse effect that catastrophically degrades the true negative rate (TNR) against out-of-chain embedders. To counteract this, a Angular Diversity Loss (ADL) is introduced that explicitly forces each new embedder to occupy a distinct region of the output space. The combined approach restores near-perfect TNR with no measurable cost to detection accuracy or perceptual audio quality.