Tolik
Log inStart
Menu

Guide

How to auto-caption a video from speech (and keep captions on the cut)

People ask “how do I add captions automatically,” “burn captions in,” and “word-level karaoke captions.” Those are not one job. A finished film that will not be recut needs a captioner or an SRT on the locked audio. A talking-head or VO-driven cut that still changes needs captions born from the same speech pass as the cut — otherwise the file lies. Template shops (CapCut, VEED) sell looks. Speech pipelines sell alignment. Tolik is the second kind.

Three caption products people confuse

File captions (SRT/VTT) you upload to YouTube: good for accessibility toggles, bad if you wanted burned-in type. Burned-in captions: part of the picture, always on, what Shorts often want. Word-level highlighting: a styling choice on top of timing, not a different transcript.

If you transcribe the raw take, then cut ums, the old SRT is already wrong. Timing has to follow the remaining words.

When a dedicated captioner is the right tool

Locked picture, legal captions, multiple languages, hearing-impaired certification. A first-pass model is not that certificate. Budget a human for anything you would defend in court or in a university accessibility office.

When a speech pipeline is enough

You still own the cut. Filler and silence may still move. You need type on the remaining words this week. The pipeline recognizes speech, drops dead air, attaches captions at word level so you fix one token (“Tolik” not “token”) without rewriting a paragraph. Podcast-length files and guest names: podcast captions. Lessons: course lesson.

Languages, names, and what models miss

Proper nouns, product names, and code fail first. Homophones fail second. Read the captions once like an intern’s transcript. That pass is cheaper than a week of comments. Do not type a script hoping the tool will lip-sync a stranger — that is overdub, not captioning.

Related cuts

Empty air should not hold a caption — remove silence. Speech driving B-roll — speech to timeline. Audio with no picture does not burn type; that is clean voiceover.

This articleNot this article
Captions that match a cut still in motionCaptioning a locked feature you will not recut
Your voice, your filesStock voice reading a blog
First pass plus a human readCertified accessibility package

Questions.

Should I upload an SRT or burn captions in?

YouTube toggle: SRT/VTT. Always-on for mute autoplay: burn in or render on the timeline.

Why do my captions drift after I cut ums?

They were timed to the uncut file. Recaption after the cut.

Are word-level captions more accurate?

They are more editable. Accuracy is still the recognizer plus your pass on names.

CapCut vs a speech pipeline?

CapCut/VEED: looks and templates. Pipeline: captions that match a cut of your files.

Is auto caption good enough for ADA/legal?

Usually no. Use a human captioner for that bar.

Two speakers?

A mixed track captions as one stream unless you bounced stems or used a dedicated diarization product.

Can I hide captions for one export?

If they live on a timeline, yes. If they are burned into pixels, re-export without them.

Do I need to type the script first?

Not for recognition captions. Typed scripts are for overdub and teleprompter jobs.

Start