Tencent Hunyuan open-sources Hy-SimulMT-30B: bidirectional CN-EN live speech translation with voice cloning
jacek2023 · reddit · 2026-10-26
Tencent Hunyuan released Hy-SimulMT-30B-A3B on Hugging Face (with GGUF builds), an open real-time simultaneous translation model doing bidirectional Chinese-English speech-to-speech translation while the speaker talks, preserving the speaker's voice via real-time voice cloning.
- Designed for meetings, talks, and interviews
- In the RealSI long-form evaluation it hit a StreamLAAL latency of 3 seconds, lowest among compared systems; its GEMBA score beat open-source rivals and four of five commercial APIs
- GitHub ships SimulInfer inference service and SimulTrain training tools; speech output requires a TTS service
- Also usable via the Hy Translate app
More from Multimodal
- Infinite Recursion Vortex prompt makes images spiral into themselves — LudovicCreator · 2026-10-26
- Dreamina offers $0.29 first month for Seedance, sparking debate on AI video subscription value — AmbitiousMeat6860 · 2026-10-26
- MiniMax amplifies workflow claim: Opus 5.5 + H3 mass-produces viral Douyin MVs in minutes — Hailuo_AI · 2026-10-26
- Midjourney sref combo --sref 1240652461 369650292 yields soft snowy illustration styles — sergeantsref · 2026-10-26
- Skip Figma for gradients — Midjourney generates stunning gradient backdrops — hewarsaber · 2026-10-26
- MiniMax H3 on 12-16GB GPUs: seven speed-up options compared — musetta_ai · 2026-10-26