Tencent Hunyuan Releases Hy-SimulMT for Real-Time Speech Translation
Tencent Hunyuan has released Hy-SimulMT, a new simultaneous translation model designed to facilitate bidirectional speech-to-speech translation between Chinese and English. According to documentation hosted on Hugging Face, the model is engineered to process incoming audio continuously, providing translated text and speech output while the speaker is still talking.
The model is specifically designed for live communication environments, including professional meetings, speeches, and interviews. A key technical feature of the system is its real-time voice cloning capability, which aims to preserve the original speaker’s vocal characteristics during the translation process.
Performance Benchmarks and Technical Specifications
In technical evaluations conducted by the developers, Hy-SimulMT demonstrated specific performance metrics regarding latency and translation accuracy. In the “RealSI” long-form audio evaluation, the model achieved a StreamLAAL (Stream Latency) of approximately 3 seconds. According to the developer’s report, this represents the lowest latency among the systems included in their comparison.
Furthermore, the model’s GEMBA score reportedly exceeded those of several compared open-source models and four out of five commercial APIs. These metrics are intended to demonstrate the model’s ability to balance translation quality with the speed required for natural conversation, thereby reducing the disruption typically caused by the delays inherent in traditional translation methods.
Availability and Implementation
The model is currently available via Hugging Face in various formats, including GGUF and Interleave-GGUF versions. For developers and researchers, Tencent has provided a GitHub repository containing the “SimulInfer” inference service and “SimulTrain” training tools. The developers note that while the model handles translation, speech output and voice cloning require a configured Text-to-Speech (TTS) service to function fully.
For end-users, the technology is accessible through the Hy Translate App.
Market Context
The release of Hy-SimulMT marks a continued expansion of Tencent’s artificial intelligence capabilities. While the technical specifics of the model are detailed in its documentation, the broader impact on the competitive landscape of AI-driven communication tools remains to be seen. The model’s performance data is based on internal evaluations and comparisons provided by Tencent Hunyuan.