OmniVoice: 600‑Language Zero‑Shot TTS Takes Center Stage

OmniVoice is a massively multilingual zero‑shot text‑to‑speech (TTS) model that supports over 600 languages, making it the broadest‑coverage TTS system currently available. Built on a diffusion language model‑style architecture and fine‑tuned from the Qwen/Qwen3‑0.6B base model, it delivers high‑quality speech with real‑time inference speeds as low as 0.025 RTF (about 40 × faster than real time). The model is released under the Apache‑2.0 license and can be accessed via the `omnivoice` Python library.

Key features highlighted in the README include state‑of‑the‑art voice cloning from a short reference audio, voice design through explicit speaker attributes such as gender, age, pitch, dialect/accent, and whisper mode, as well as fine‑grained control using non‑verbal symbols (e.g., `[laughter]`) and pronunciation correction via pinyin or phonemes. The repository provides a Colab notebook, a Hugging Face Space demo, and detailed usage instructions for generating speech in zero‑shot mode or with custom voice designs.

The model’s multilingual capability, combined with its fast diffusion‑based inference and flexible control mechanisms, positions OmniVoice as a strong foundation for applications ranging from multilingual audiobooks and real‑time dubbing to personalized voice assistants and low‑latency TTS services.