System embeds resilient watermarks in synthetic speech to trace origin
Learning to Watermark Speech Synthesis Against Model-Driven Reconstruction
Cryptography and SecuritySound
Summary
Synthesized speech can be copied and altered, making it hard to tell where it came from or who made it. The authors developed a method called Thrive that hides special codes inside computer-generated speech, which can survive changes made during editing and sharing. This lets people check if a speech clip came from a specific text-to-speech system and which user requested it. Their tests showed Thrive works well across different types of speech systems and tough editing scenarios, recovering identifiers accurately.
What this means in practice
- •For tts service providers: Maintain reliable user-linked watermarks in synthesized speech to verify speech origin even after post-processing and editing.$Commercial implications: Enables sale of watermarking tools that guarantee content provenance and copyright protection in large-scale TTS deployments.
- •For audio forensics teams: Identify the source of synthetic speech samples after potential modifications using robust watermark recovery under diverse transformations.
Authors
Weizhi Liu, Yue Li, Hui Tian, Zhaoxia Yin
Abstract
Modern TTS systems increasingly generate synthetic speech at scale for diverse users. This setting calls for content-level provenance that can verify the origin of released speech and attribute it to the requesting user, which generative watermarking can support by embedding multi-bit identifiers directly into synthesized speech. Once released, however, speech may undergo heterogeneous learned transformations during distribution and editing, with reconstruction objectives that can preserve speech utility while affecting watermark recoverability differently. We find that no single watermark carrier remains consistently reliable across reconstruction models, as its survival depends jointly on the embedded structure, reconstruction mechanism, and observation representation. To this end, we propose Thrive, a multi-bit generative speech watermarking framework for modern autoregressive TTS, covering both discrete-token and continuous-representation generation under reconstruction attacks. Specifically, Rise synchronizes watermark injection into intermediate representations with its continued integration into subsequent generation, while Care combines waveform and spectral experts using bit-wise reliability selection. Experiments on both autoregressive paradigms show that Thrive preserves synthesis fidelity, achieves 87.6% average recovery accuracy under reconstruction attacks, and supports source attribution over candidate sets of up to 10,000 identities.