Correcting the transcript
The Review step of Auto captions - fix a word, adjust the timing, hide a caption, name the speakers, replace a term everywhere and render again.
No caption is burnt in before the export: until then, the transcript stays a track you can correct freely. The Review step of Auto captions is there for that. It is optional, but a few minutes are enough to clear the uncertain words and check the voice colors.
Opening the Review step
- Transcribe at least one video in the Transcribe step; otherwise the Studio shows Transcribe at least one video so it can be corrected.
- Click Review the captions.
- The Videos in the batch strip lists your videos; pick the one to correct. The summary shows how many uncertain words are left across the batch.
The page shows the video with a caption preview identical to the export, and the list The video's captions. The Full screen, captions included button lets you judge the result at full size.
Finding what deserves a second look
The All, Changed, Hidden and Uncertain filters narrow the list. A word tagged uncertain is one the transcription doubts; the Needs review first option moves those captions to the top. The Search the transcript field finds a precise sentence. Clicking a caption seeks the video to its moment.
Correcting a caption
- Click into the Caption text and fix the word. The preview updates at once.
- Adjust the start or the end if the caption comes too early or stays too long.
- Hide this caption takes a line out of the render without deleting it; Restore caption puts it back.
- Undo the corrections on this caption returns that line to the original transcript.
Captions fit on a single line: a lengthened sentence is re-split at render time rather than spread over two lines.
Replacing a term across the batch
A first name, a brand or a technical word that was mis-transcribed often comes back in several videos. Type the word in the search field, the replacement under With, then click Replace everywhere. The counter shows how many replacements were made.
Speakers and their colors
When speaker detection is on, each voice gets a color from its timbre. For each speaker:
- Name of speaker: a name to find your way around the list.
- Voice of speaker: Female voice, Male voice or Uncertain voice. Changing this role recolors every line of that voice.
If a voice stayed unresolved, the tool says so at the end of the transcription (N unidentified voice: listen to the samples…). In the Match the speakers section of the Transcribe step, an audio sample per voice helps you set the right color. A caption attributed to the wrong person is fixed by changing its Speaker.
After the corrections
- If the render has not started yet, continue to Export: the corrections are taken into account directly.
- If the render is already done, the export shows Captions changed since this render: render it again to include it in the archive. Click Re-render the changed videos; only the changed videos are rendered again, then Download the up-to-date files (ZIP).
The caption position cannot be changed here: it stays the one confirmed for the batch in the Configure step.
If the transcript is wrong everywhere
Check the Transcription language: an English video transcribed as French gives incoherent words. Changing the language transcribes the affected videos again, and that is paid for again; the tool tells you beforehand (Redo N transcriptions?). Your current corrections are kept aside and come back if you restore the previous setting.