
Check the Media source choice in Express's caption generator. The editor offers Videos, Audio and All. If the spoken words are on a separate voiceover track, generating from video sources alone can be the wrong request. Choose the relevant source and review the resulting transcript against the actual soundtrack.
Key Takeaways
- Establish where the speech lives before generating captions.
- Media-source selection and spoken-language selection are different settings.
- Generated text needs a final words, timing and placement review.
Locate the speech in the project
Listen to the video source and the added voiceover separately using the available track controls. Identify which contains the approved speech. A picture can have music while its explanation comes from another file, so “the video has sound” is not enough to specify a caption source.
Adobe's Caption videos instructions document Open captions, the spoken-language dropdown and the Videos, All and Audio source choices. Use those choices to reflect the project you actually assembled.
If the only audible material is music or effects, caption recognition cannot supply dialogue that was never spoken. The speech-source diagnostic guide covers that different issue. Repeated generation is not a substitute for adding the intended narration.
Generate from the approved content
Save the current project version and confirm that the correct voiceover revision is present. Open captions, choose the language actually spoken and select the appropriate media source. If the project contains competing speech tracks, review which material should contribute rather than assuming All is always the cleanest choice.
Generate the captions and inspect the Transcript panel. Correct names, numbers and technical terms against the accepted script and the audible performance. The language setting describes source speech; it is not a promise to translate the explanation into a different language.
Review the picture at the same time. Captions should not cover essential product text or a face. Adobe documents formatting, repositioning and resizing on the canvas; use those controls to achieve readable placement without changing the spoken meaning.
Check after the soundtrack is final
For a hypothetical silent generated demonstration, an editor adds a separate explanation and a low-level score. Generate the candidate captions from the intended spoken source, then check the opening sentence, a distinctive product name and the final instruction.
The proposed acceptance conditions are that the words match the approved narration, the text appears with the relevant speech and no rejected take is captioned. If the voiceover is replaced or shifted later, review the captions again. A correct transcript from an earlier sound revision can become wrong without any spelling mistake.
Export and inspect the actual delivery too. Captions that look good in the editor are not proof that every downstream format supports the same styling. The SRT destination guide explains that separate compatibility boundary. Preserve the final captioned revision alongside the accepted audio so the published words remain traceable to the soundtrack viewers actually hear.
Related tasks
- Why Does Caption Recognition Say “No Audio” When Background Music Is Playing?
- Why Does Premiere Export the Wrong Language When Two Caption Tracks Are Enabled?
- Why Does My Styled SRT Lose Its Color After Upload to YouTube?
Sources and scope
- Caption videos: Timeline Open captions > spoken language > Media source Videos/All/Audio > Generate.; Edit text in Transcript; Format styles captions; move/resize on canvas.
- Add audio tracks to designs: 3 tracks maximum:2music/effects,1voiceover.; Add content > Media > Audio; upload,stock,Record voiceover.; Nonactive tracks collapse; select to expand; handles duration; menu split/mute/delete.
Documentation-based workflow with proposed review criteria.
Prepared by Imagild Editorial, the company editorial team responsible for these guides. AI assisted research and drafting. This is documentation-based guidance with proposed examples, not a claim of hands-on testing or an independent product review. Report a correction.
Cover: AI-generated conceptual editorial illustration; not a real application screenshot, observed output or before-and-after test.