Tencent open-sources Hy-SimulMT: real-time speech-to-speech translation with ~3s latency
aigclink · x · 2026-10-26
Tencent Hunyuan has open-sourced Hy-SimulMT, a real-time simultaneous translation model supporting bidirectional Chinese-English speech-to-speech translation with a streaming latency (StreamLAAL) of about 3 seconds, approaching professional interpreters in some tests.
Key capabilities:
- Streaming interpretation: outputs translation while the speaker is still talking, rather than waiting for sentence boundaries
- Low latency: optimizes simultaneous-translation decisions for continuous speech — when to translate, how much, and how to handle unheard context
- Real-time voice cloning: pairs with TTS to generate target-language speech preserving the original speaker's voice
The 30B-A3B model ships with BF16, FP8 and GGUF weights, plus a SimulInfer inference service and SimulTrain training pipeline. It targets meetings, talks, interviews and live streams, outperforming the traditional segmented ASR → MT → TTS pipeline in latency and fluency.
More from Models
- Model refuses wrong psychiatric advice in deliberately misleading test — FellMentKE · 2026-10-27
- Clinical System One v0.2: top clinical AI drops from 97% to 65% on harder cases — MaziyarPanahi · 2026-10-27
- ChatGPT keeps telling people 404 Media is tech support for big corporations — 404 Media · 2026-10-26
- Claude's /show-me visualizes compressor auto makeup gain like a pro audio teacher — teropa · 2026-10-26
- ChatGPT keeps telling people 404 Media can fix their tech support problems — 404 Media · 2026-10-26
- Fern's "context is intelligence" thesis: prediction keeps improving out to 10.2M tokens — bneyshabur · 2026-10-26