MetaVoice

MetaVoice-1B is a 1.2 billion parameter foundational TTS model trained on 100K hours of speech for human-like, expressive text-to-speech synthesis. It prioritizes emotional speech rhythm and tone in English, with zero-shot voice cloning support for American and British voices using just 30 seconds of reference audio. MetaVoice also supports cross-lingual voice cloning with fine-tuning — requiring as little as 1 minute of training data. The model can synthesize arbitrary-length text and comes with a Docker-based Web UI and API server. Key features include: emotional speech rhythm and tone, zero-shot cloning for US/UK voices, cross-lingual voice cloning with fine-tuning, arbitrary-length text synthesis, Docker Compose deployment with Web UI, and Apache 2.0 license for unrestricted use.