Captions in sync with every word

Word drift is not a whole-file offset

If every caption is late by the same amount, the whole file is shifted and one number fixes it; the free subtitle shift tool does that for SRT and VTT files. Word drift is different: some words are early, some late, and the errors gather around pauses.

Speech models estimate where each word begins and ends. Next to a pause those estimates often reach into the silence, so a word appears before it is heard or stays lit after it is finished. With word-by-word highlighting, even a tenth of a second shows.

What the automatic snap does

Every transcription in Cuecaps, on-device or cloud, ends with a timing pass that reads the loudness of your audio in 10 ms steps.

• A word edge next to a silence moves to where the sound actually rises or falls, up to 0.15 seconds away.
• Edges between words said in one breath stay where they are, since there is no silence to snap to.
• A silence has to last at least 0.08 seconds, so the brief dip of a 't' or 'p' inside a phrase does not count.
• Words never overlap and are never shortened below 0.06 seconds.
• If music or steady noise sits under the voice, the pass changes nothing rather than guess.

After editing words you can run it again: press Ctrl+K (Cmd+K on a Mac) and choose Fix timing. Undo puts the previous timing back.

Fine-tuning on the waveform timeline

1. Play the clip and watch the highlight. A karaoke-style highlight makes timing errors easy to spot.
2. On the timeline under the preview, drag a word block to move it, or drag one of its edges to trim it.
3. Edges snap to the neighbouring word and to the playhead, with a guide line showing the snap, so words line up without gaps.
4. Use Start a new caption after a pause of, in the Layout tab, to control when a pause clears the screen.

On a phone, the timeline lives in the Timing tab, so the preview keeps the screen while you watch.

Questions