Production-ready text-to-speech for dubbing, games, podcasts, and education — precise duration control, emotion–speaker decoupling, zero-shot cloning