Latent-Mark: An Audio Watermark Robust to Neural Resynthesis

2026-03-05

AI Summary

This work proposes the first zero-bit audio watermarking framework designed to be robust against semantic compression, such as that induced by neural audio codecs, which often removes subtle waveform perturbations used by conventional methods. By introducing directional perturbations in a shared, codec-invariant latent space of neural audio codecs, the watermark manifests as a detectable shift in the direction of latent representations while remaining imperceptible to human listeners. The approach integrates cross-codec joint optimization with audio manifold constraints to achieve zero-shot transfer robustness against unseen neural codecs. Experimental results demonstrate state-of-the-art detection performance under diverse neural resynthesis attacks and traditional digital signal processing operations, all while preserving high perceptual audio quality.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

While existing audio watermarking techniques have achieved strong robustness against traditional digital signal processing (DSP) attacks, they remain vulnerable to neural resynthesis. This occurs because modern neural audio codecs act as semantic filters and discard the imperceptible waveform variations used in prior watermarking methods. To address this limitation, we propose Latent-Mark, the first zero-bit audio watermarking framework designed to survive semantic compression. Our key insight is that robustness to the encode-decode process requires embedding the watermark within the codec's invariant latent space. We achieve this by optimizing the audio waveform to induce a detectable directional shift in its encoded latent representation, while constraining perturbations to align with the natural audio manifold to ensure imperceptibility. To prevent overfitting to a single codec's quantization rules, we introduce Cross-Codec Optimization, jointly optimizing the waveform across multiple surrogate codecs to target shared latent invariants. Extensive evaluations demonstrate robust zero-shot transferability to unseen neural codecs, achieving state-of-the-art resilience against traditional DSP attacks while preserving perceptual imperceptibility. Our work inspires future research into universal watermarking frameworks capable of maintaining integrity across increasingly complex and diverse generative distortions.

Problem

• audio watermarking
• neural resynthesis
• semantic compression
• neural audio codecs
• robustness

Innovation

• audio watermarking
• neural resynthesis
• latent space
• zero-bit watermarking
• cross-codec optimization