What music does to auto captions
A speech model has to pick the voice out of everything else in the track. Quiet music under clear speech rarely matters, but a loud beat, a drop or sung lyrics right under a sentence can turn words into near-misses, and a song with vocals can end up in your captions as if you had said it.
Timing suffers more quietly. After transcription, Cuecaps moves word edges onto the short gaps between words, and it finds those gaps where the loudness drops. With music filling every pause there is nothing to find, so the studio keeps the model's own timing rather than guess, and some words light up a little early or late.
Best: caption the voice before the music
- In your editor, mute the music tracks and export the voice on its own, at the same length as your final edit.
- Drop that file into the Cuecaps studio and run Find words. Clear speech is what On this device does best, so it can run free in your browser.
- Pick a style and read through the words. Word timing snaps cleanly now, because the pauses are real silences.
- Export a transparent overlay (MOV or WebM, on Creator and Pro), or SRT and VTT files on any plan.
- Put the overlay on the track above your final edit, starting on the same frame, and export the finished video with the music.
Since the voice file and the final edit start on the same frame and run the same length, the captions line up without nudging.
When you only have the mixed file
- Choose Cloud in Find words: the studio marks it as the most accurate for timing. It needs a free account, which includes 10 cloud minutes a month.
- Type names, brands and slang into Names and terms to spell correctly, so the model doesn't have to guess them over the beat.
- If song lyrics turn up as captions, select those words and press Del.
- Play the clip and watch the highlight. Where a word is early or late, select it and press [ or ] at the right moment to set its start or end to the playhead.
- For finer nudges, Alt+Left and Alt+Right move the selected word by 10 ms; add Shift to move it by 100 ms.
Fix timing will mostly leave a mixed track alone, for the same reason the first pass did, so hand timing is the tool here.
Questions
Why are my auto captions wrong when there's music?
The speech model hears the music as part of the audio, so a loud beat or sung lyrics can bury spoken words or add words nobody said. Captioning a voice-only export avoids it.
Can Cuecaps remove the music before transcribing?
No. The studio doesn't split voice from music. Export the voice on its own from your editor, or use cloud transcription on the mix and correct what it misses.
Why don't my captions snap to the words when music plays?
Automatic snapping looks for quiet gaps between words. Music fills them, so Cuecaps keeps the transcribed timing rather than snap to the wrong place. Set those words with [ and ] or drag them on the timeline.
Can I caption a song's lyrics?
Cuecaps is made for speech, and speech models are less reliable on singing, so expect to correct lines by hand with Edit word (Enter).
Try it on your video
You can also drop a file anywhere on this page. Or tryTry the demo clip