Tencent Hunyuan releases Hy-SimulMT, a low-latency real-time Chinese-English simultaneous translation model
Tencent Hunyuan released Hy-SimulMT, a real-time simultaneous translation model for bidirectional Chinese-English speech-to-speech translation with voice cloning. It achieved a StreamLAAL of approximately 3 seconds, the lowest among compared systems, and its GEMBA score exceeded four of five commercial APIs. The repository provides SimulInfer and SimulTrain tools; speech output requires a configured TTS service.
China context
- Original name
- 腾讯混元
- Outside China
- Open weights · github.com
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China can access the model weights and inference/training tools from the GitHub repository, enabling integration into real-time translation applications. However, speech output and voice cloning require a configured TTS service, which may add setup complexity.
- For investors
- The release of a competitive open-source simultaneous translation model by Tencent Hunyuan may pressure commercial API providers in the real-time translation market, potentially affecting pricing and differentiation strategies.
Translated from Chinese. Quotes and facts link to the original sources.
Hy-SimulMT is a real-time simultaneous translation model developed by Tencent Hunyuan, supporting bidirectional speech-to-speech translation between Chinese and English. It continuously receives incoming audio and translates as the speaker talks, producing both translated text and speech, with real-time voice cloning to preserve the speaker's vocal characteristics. The model is designed for live communication settings such as meetings, speeches, and interviews. In the RealSI long-form audio evaluation, Hy-SimulMT achieved a StreamLAAL of approximately 3 seconds, the lowest among the systems shown, and its GEMBA score exceeded those of the compared open-source models and four of the five commercial APIs. The repository provides the SimulInfer inference service and SimulTrain training tools. Speech output and voice cloning require a configured TTS service. The model can be tried in the Hy Translate App.
Hy-SimulMT uses a streaming architecture that processes continuous audio input and generates incremental translations, balancing latency and quality. The reported StreamLAAL of ~3 seconds indicates low streaming lag, and the GEMBA scores suggest competitive translation quality. The repository includes SimulInfer for inference and SimulTrain for training, but details on model size, architecture, and training data are not provided in the evidence.
Developers and enterprises needing real-time cross-language communication gain access to an open-source model that claims lower latency than existing open-source and most commercial APIs, potentially reducing the cost and complexity of building live translation services. Competitors offering commercial simultaneous translation APIs may face pressure if Hy-SimulMT's performance is independently verified.
Hy-SimulMT provides a foundation for building real-time translation applications in meetings, speeches, and interviews, potentially reducing latency and improving user experience compared to existing solutions. The open-source release allows integration and customization, but commercial deployment may require additional TTS infrastructure and licensing considerations.
Independent evaluation of Hy-SimulMT on public benchmarks and real-world audio will be a key signal to watch. The release of training tools (SimulTrain) may enable further research and fine-tuning. Adoption in the Hy Translate App and any subsequent updates to the model or repository will indicate ongoing development.