Word-level caption editing

Word-level caption editing: fix, bleep, cut and highlight single words.

Word-level caption editing means the word is the unit you work with, not the line. CaptionFX keeps a stable identity and real timing on each word, so you can correct one, censor one, remove one from the video entirely, or emphasise one - without disturbing anything around it.

Short answer

In CaptionFX every word has its own identity and timing, so you can correct one word, bleep it, cut it out of the video or highlight it without disturbing the rest of the line. Nothing changes your original upload, and edits apply only when you export.

Last updated

Bleep one word

Cover a single word's audio with a tone or silence while the video runs on. The rest of the sentence is untouched.

Cut one word

Remove a word from the audio and the video, then re-time every caption after it so the whole clip still lines up.

Highlight one word

Emphasise the word a line turns on, either from the automatic pass or chosen by hand where you know better.

Why does the word have to be the unit?

Most caption editors work on lines. You get a block of text with a start and end time, and you edit the block. That is fine for fixing a typo and useless for everything else, because the things you actually want to do to a video are word-sized. One swear word. One filler. One name said wrong. One stumble. One word that should land harder than the rest.

If the line is your unit, each of those becomes a bigger operation than it should be, and some become impossible. You cannot bleep a word if the smallest thing you can address is the sentence containing it. You cannot cut a stumble without cutting the phrase around it.

CaptionFX makes the word the unit. Each word carries its own timing and its own identity, and operations attach to the word rather than to a position in a line. The identity part matters more than it sounds: it means an effect you put on a word stays on that word even after you edit the text around it, insert something earlier, or re-time the line. Without stable identity, effects are stored by position, and editing earlier in the sentence silently slides your bleep onto a different word.

What is the difference between bleeping and cutting a word?

Bleeping replaces a word's audio - a tone, or silence - and leaves the clip's length alone. The video keeps running, the speaker's mouth keeps moving, and the word is simply not audible. Use it when the timing of everything else matters, or when you want it obvious that something was censored.

Cutting removes the word from the audio and the video. The clip gets shorter, and everything after the cut moves earlier - which means every subsequent caption has to be re-timed, or the whole rest of the video is out of sync. CaptionFX does that re-timing as part of the cut. Use it for stumbles, false starts, long filler words and anything you want gone rather than hidden.

Both are applied at export, on a copy. Your source video and the caption text are never modified, so an effect you change your mind about is removed by removing the effect - not by undoing damage. You can export, dislike it, drop the cut, and export again from the same original.

Why can some words not be bleeped?

A word can only anchor an audio edit if its timing is real. Speech models measure timings for most words directly from the audio, but not all - some are interpolated from the words around them, and a word whose text you have rewritten no longer corresponds to a measured sound at all.

Acting on an estimated timing produces an edit that misses. A bleep that starts a fraction late leaves the first syllable of the censored word audible, which defeats the entire purpose and is worse than having done nothing. A cut on a guessed boundary clips the end of the previous word.

So CaptionFX tracks which words have measured timing and which do not, and only lets measured words carry a cut or a bleep. If a project has an effect on a word whose timing is no longer trustworthy, the export is refused with an explanation rather than rendering something subtly wrong - and it is refused before you are charged, not after. It is a deliberately conservative rule: you are told to repair the timing, because a censor that half-works is a failure that reaches your audience.

Do corrections destroy the original transcript?

Rewriting a caption line is lossy in the obvious implementation. You replace the text, and the original words - with their measured timings, the provider's original tokens, the speaker information - are gone. Later you want to cut a word in that line and discover you no longer have the timing to do it with.

CaptionFX keeps what an edit drops. When a correction removes words from a line, the original measured words are retained alongside rather than discarded, so a caption fix never destroys underlying transcript data. The same applies to translation: translated captions have no word-level timing of their own, so the measured source words travel with them instead of being thrown away.

The practical effect is that correcting text early does not close off options later. You can fix the wording now and still cut a word from that line afterwards.

How it works

  1. Generate captions with word timingTranscription produces timings for individual words, and records which of them were measured from the audio rather than estimated.
  2. Correct the textFix wrong words in place. The original measured words are kept alongside the correction rather than discarded.
  3. Bleep or cut specific wordsSelect the word and choose a tone, silence, or removal. A cut re-times every caption after it so the rest of the clip stays in sync.
  4. Set emphasis where it earns itAccept the automatic choice or pick the words yourself. One or two per line - beyond that, emphasis stops distinguishing anything.
  5. ExportEffects are applied at export on a copy: cuts first, then the caption burn-in, then bleeps. Your source video is never modified.

Questions people ask

Can I bleep a single word in a video?

Yes. Select the word and choose a tone or silence. The clip's length does not change - the video keeps running and that word is simply not audible. Only words whose timing was measured from the audio can be bleeped, because an estimated boundary produces a bleep that misses.

What is the difference between bleeping and cutting a word?

Bleeping replaces the audio and leaves the clip the same length. Cutting removes the word from both audio and video, making the clip shorter, and every caption after the cut is re-timed so the rest stays in sync. Bleep to censor, cut to delete a stumble or filler.

Why can I not bleep a particular word?

Because its timing is estimated rather than measured from the audio - usually a word the speech model interpolated, or one whose text you rewrote. Acting on a guessed boundary leaves the first syllable audible, so CaptionFX refuses rather than shipping a censor that half-works, and it tells you before you are charged for the export.

Can I highlight selected caption words?

Yes. CaptionFX can choose emphasis words automatically, and you can override that for any line by hand. Keep it to one or two words per line - if most of a line is highlighted, the highlighting no longer distinguishes anything.

Does cutting a word break the rest of my captions?

No. A cut shortens the clip, which would put everything after it out of sync, so re-timing the subsequent captions is part of the cut rather than something you do afterwards.

Does editing captions change my original video?

Never. Every effect is applied at export, on a copy. The caption text, the word timings and the uploaded video are left intact, so removing an effect genuinely undoes it and you can export again from the same source.