Streaming Speech-to-Text Translation with Open-LiveTranslate

Instructions:

Select a language pair and chunk size
Click the microphone to start recording
Refresh the page to reset the translation history
If you experience no response or slow response, please refresh — the serving engine (vLLM) may be experiencing a cold start

Language Pair

Select source → target language.

Chunk Size (seconds)

Larger chunks = more context but slower response. Must be multiple of 0.96s.

Audio Input

Backend Status

Translation

Note: Currently supports English → Chinese translation only. This model is trained with the method described in the ACL 2025 paper InfiniSST: Simultaneous Translation of Unbounded Speech with Large Language Model by Siqi Ouyang, Xi Xu, and Lei Li.