tencent/Hy-SimulMT-30B-A3B-Interleave-FP8
tencent/Hy-SimulMT-30B-A3B-Interleave-FP8
Tencent Mixue released the Hy-SimulMT real-time simultaneous translation model, supporting bidirectional Chinese-English voice-to-voice translation, and can continuously output translated text and speech during speaking, while retaining the speaker's voice through real-time voice cloning. In the RealSI long audio evaluation, its StreamLAAL is about 3 seconds, the lowest among compared systems, and the GEMBA score is higher than the open-source models and four out of five commercial APIs being compared.
The official provides a comparison of latency and quality for real-time simultaneous translation, and opens the model weights and inference deployment scripts, which can be used to evaluate deployment methods.
Source: Tencent Hunyuan · Hugging Face new models · huggingface.co