Qwen3-TTS (3-second cloning), F5-TTS, XTTS, RVC and Chatterbox lower the cost of cloning. Chatterbox embeds a PerTh watermark and AudioSeal offers an MIT-licensed one, while VibeVoice withdrew its TTS code after misuse. Which combination of consent record, watermark and detector is practical, robust to re-encoding, and enforceable across open models whose licenses differ?
Why this is still open: Watermark robustness is claimed by authors; no cross-model consent+provenance standard was found.