The project accepts streaming speech and produces translated text and speech while the speaker is still talking.
The repository describes bidirectional Chinese-English speech-to-speech translation and real-time voice cloning intended to retain a speaker's vocal characteristics. It includes inference and training components, plus deployment documentation and demos. Tencent links the release to Hugging Face and ModelScope collections.
The Hugging Face collection lists six variants, including a 30B-parameter mixture-of-experts configuration with 3B active parameters, offered in FP8 and GGUF formats. The supplied source does not establish independent translation quality, latency under a defined hardware setup, voice-cloning safeguards, production availability, or support for languages beyond Chinese and English. Source: https://github.com/Tencent-Hunyuan/Hy-SimulMT