On October 26, Tencent Hunyuan's simultaneous interpretation large model Hy-SimulMT was released and open-sourced, supporting real-time Chinese-English speech-to-speech translation: the speaker's audio can be input to the model in real time, translated while listening with simultaneous output of translated text and speech, and real-time voice cloning is supported, reproducing the speaker's voice and tone. The model debuted for free experience on Tencent's Hy Translation App, with the open-source repositories simultaneously launched on HuggingFace, ModelScope, and GitHub.
Low latency is the model's core metric, with measured response at about 3 seconds, lower than the 4-5 second average of human professional interpreters. In comparisons with 6 certified simultaneous interpretation experts (3 CATTI Level 1 and 3 CATTI Level 2) on audio from ten domains including software AI, medicine, manufacturing and energy, and business management, Hy-SimulMT achieved a StreamLAAL latency of 3.3 seconds, better than the CATTI Level 1 experts' 4.5 seconds and Level 2 experts' 4.7 seconds, with translation quality reaching the level of human simultaneous interpretation experts.
On the long-audio evaluation set RealSI, StreamLAAL measures latency and GEMBA measures translation quality (scored by GPT6.1-Sol); Hy-SimulMT achieved about 86 points with a latency of about 3.1 seconds, with the lowest latency and the best quality among open-source models, surpassing most commercial simultaneous interpretation APIs. In the four short-audio tests, its RealSI-Sentence score of 84.7 was the highest overall, and FLEURS90.3 and ACL606081.2 ranked first among open-source models, while Gemini3.5-LiveTranslate ranked first in individual tests with FLEURS92.4 and ACL606082.6.
The model controls latency with a "speak first, then revise" strategy, with revisions sounding close to an interpreter's self-correction. Wang Weiwei, Vice Dean of the School of Interpreting and Translation Studies at Guangdong University of Foreign Studies, and Professor Zhong Weihe, former Executive Vice President of the Translators Association of China, believe that its speech recognition stability, simultaneous interpretation response speed, and interpretation delivery quality all perform impressively, advancing machine simultaneous interpretation into real-world scenarios.