No delay
The track is translated in batches ahead of the video, so each line is ready before it is spoken.
When a video already has subtitles, SubLive does not need to listen to it. It takes the subtitle track, translates it ahead of playback and shows each line at the moment it is spoken. Pause, rewind or watch at double speed, and the translation stays in step.
Kelios minutės nemokamai patvirtinus el. paštą. Be kortelės.
What is said
Translated from the sound
Translated from the subtitle track
The track is translated in batches ahead of the video, so each line is ready before it is spoken.
Auto-generated YouTube captions arrive word by word. SubLive joins them into sentences and translates each one with the lines around it.
SubLive translates about ninety seconds of the track ahead of playback and asks for the next stretch thirty seconds before that runs out, so the video never waits for the translator.
On YouTube, or on a site whose HTML5 player has a subtitle track, such as TED. Start SubLive the usual way.
Human-made subtitles come before auto-generated ones. If a human track in your language already exists, it is shown as it is, with no machine translation.
One switch in the extension makes SubLive listen to the sound instead. Videos without subtitles, live streams and calls always work that way.
When SubLive translates from the sound, the clock starts only once a phrase has been said. It waits 0.7 seconds of silence to be sure the phrase is over, translates it in about three quarters of a second and puts it on screen. The AI voice needs roughly another second before its first sound.
From the subtitle track, all of that happens ahead of time. The line is already translated when the speaker starts the phrase, so it appears together with it: zero seconds after the phrase begins, not 1.6 seconds after it ends. Only the very start of a video waits once, while the first stretch of the track is translated.
Seconds from the end of a phrase until you see or hear it
Each step in seconds, on one scale: the full circle is three seconds.
Subtitle, translated from the sound
AI voice, translated from the sound
Subtitle or voice from the subtitle track
Translation is the average over 386 phrases on SubLive servers on 20 October 2026. The pause and the display step are fixed settings, and the first sound of the AI voice is our own measurement. Recognition can add a little on top, while translation sometimes starts before the pause is over, so the totals are approximate.
Šaltiniai: SubLive production metrics, 20 October 2026; speech recognition setting: 700 ms of silence ends a phrase; SubLive latency review: about 1.1 s to the first sound of the AI voice
Yes, in every language YouTube generates them for. Automatic captions come in short fragments, a word or two at a time, so SubLive first puts the words back into sentences and only then translates them. Whenever a video also has human-made subtitles, those are used first.
The AI voice can read the translated subtitles aloud. Each line starts together with the speech it replaces, and the video's own sound gets quieter underneath.