Real-time translation for live streams
TH-Labs translates and re-voices a live stream as it happens — about two seconds behind the speaker, in their own cloned voice, in 40+ languages at once.
Live dubbing, not just live captions
Live captions make a multilingual audience read along. TH-Labs gives each viewer an audio track in their own language, spoken in the presenter's voice, so they can simply watch and listen.
The pipeline runs continuously: speech is recognised, translated and voiced in a rolling window, keeping end-to-end delay at roughly two seconds.
Where it is used
Conferences and panels with an international audience, webinars and online lectures, product launches and broadcasts. Several target languages can run from the same stream, and subtitles are produced alongside the audio.
Developers can connect streams programmatically through the TH-Labs API.
How live translation works
- 01Connect your live stream.
- 02Choose the target languages.
- 03TH-Labs transcribes, translates and voices the speech continuously.
- 04Viewers hear the stream in their language about two seconds behind the original.
Frequently asked questions
- How much delay does live translation add?
- About two seconds end to end.
- Can one stream be translated into several languages at once?
- Yes. Pick as many of the 40+ supported languages as you need.
- Does the translated voice sound like the speaker?
- Yes. The speaker's voice is cloned from about three seconds of clean audio and used for every language.