Captions on webinar clips after the cut

Caption the clip, not the webinar. The insight is a range. The VOD is a mix: two mics, slide rustle, “can you see my screen,” a Q&A that is not the post. Automatic speech recognition on that mix is a draft, not a finished track. Cut first. Identify speakers. Then burn.

The VOD is inventory

A webinar recording is a source file. A clip is one claim: a definition, a number, a story that stands without the previous forty minutes. Feeding the whole file to a captioner typesets queue talk, dead air, and every “next slide.” You then discover the sentence you needed lived at 00:41:12.

Trim until the speech you will post is most of the file. That is the same order as any long source: isolate the keep, then caption the keep. A Zoom mix degrades ASR on purpose. Overlap, compressed audio, a laptop fan, a second speaker coming in hot — the recogniser will swap words and drop names. Four minutes of that is recoverable. Fifty-five minutes is a transcript you will not read.

If the hook is a line that was never said — a title card, a takeaway you wrote in the recap — that is an overlay on the clip, not a caption hope. The captioner transcribes. It does not summarise the webinar.

Picture lock includes the crop. A 16:9 webinar dumped into 9:16 will put the speaker in a postage stamp and the captions in the chrome. Reframe, then caption. Recaptioning after a vertical crop is a second bill for a first-order mistake.

Mute autoplay is why the clip needs words on screen. That is an argument for captioning the short. It is not an argument for captioning the archive.

Multi-speaker needs IDs

Webinars are rarely one voice. Host, guest, moderator, a question from the floor that the laptop mic half-caught. A speech-only block with no speaker label is a mute viewer’s guess. On a talking-head Reel you can see the mouth. On a two-box Zoom you often cannot tell who owns the line, especially after a crop that kept only one face.

Speaker IDs are transcript convention, not a lower third. ALEX: and PRIYA: at the start of a cue, or a short name on the first line of a two-line block. Use them when:

two voices share the mix
the speaker is off screen
the clip cuts between boxes faster than a scroller can track faces

Do not ID a single on-camera host for the entire clip. Redundant labels spend CPS. Do not confuse IDs with a name super. The super is a graphic you hold for two seconds. The ID is who owns this sentence.

ASR will not reliably label speakers on a Zoom mix. It will also merge overlap into one nonsense clause. If two people talk at once, pick a winner or split the clip. Captioning crosstalk is how a price and a caveat land as one sentence.

SDH is a further pass: [laughter], [slide change], [music] when those sounds are the point. Most webinar clips do not need a full SDH track. If the joke is a door, write the door. If the joke is the sentence, caption the sentence.

Language code matches the spoken language, one code per range. A bilingual Q&A is two ranges or authored mixed lines. 165 codes on add_veed_captions is a listen list, not a simultaneous interpreter.

ASR on the mix is a draft

Treat the first render as a transcript to edit, not a ship file.

Names. Guests, companies, products. The reason you booked the webinar is the string ASR will phoneticise. Build a list from the run-of-show before you caption.

Numbers. Dates, prices, percentages, “three things.” A muted clip that gets the figure wrong is worse than no clip.

Slide text versus speech. Presenters read decks, paraphrase decks, and contradict decks. Captions follow the mouth, not the slide. If the slide is the source of truth, you are making a different object — a text overlay of the takeaway — and you should not pretend the ASR is that overlay.

Hold music, recording disclaimers, “we’ll wait thirty seconds.” Trim them. Captioning the disclaimer is how your first two seconds on TikTok are a legal paragraph the organic viewer did not come for. Paid clips of the same range may still need a disclosure; that is an overlay on the ad render, not a reason to leave the Zoom pre-roll in the organic file.

Read the transcript on the clip, not on the VOD. You will actually finish a 40-second read. You will not finish a 55-minute one, and the miss will be the guest’s name.

BASIC (21 looks at 1×) is the default for webinar clips that will also live as course snippets. DYNAMIC (9 at 2×) only if the clip is a hook Reel and the mute pass is still readable. Word-pop on a dense definition is a strobe. Two-line blocks are what an insight cut can hold.

Cut, ID, burn

Order:

Mark the insight. One claim.
Trim. New file.
Reframe to the destination. New file if the ratio changed.
Set the language code to what was spoken.
Transcribe. Add speaker IDs where the face does not do the work.
Spot-check names and numbers against the run-of-show.
Burn. Preview the style on the first seconds. Watch on mute at phone size.

The AI caption generator and add captions to a video are that burn. They will not find the insight in the VOD. Add subtitles automatically is the same engine with an accessibility-first BASIC intent — still on the clip, still after the cut.

If you need a sidecar for the course player, export the edited transcript as SRT or VTT for the long version. Convert in the VTT to SRT converter rather than renaming files. That sidecar is not the TikTok. The TikTok is pixels.

A webinar clip that ships with a wrong guest name, no speaker ID on a two-box, and karaoke on a 40-word definition is three failures that all trace to captioning before the cut was a clip. Make the short. Then caption the short.

Clip first, then burn in Versely.

FAQ

Should I caption the full webinar then cut clips?

No. You will pay to transcribe dead air and then throw it away. Trim the insight. Caption that file. A later trim of a burned clip is a recaption.

Does auto-captioning label host and guest?

Not reliably on a Zoom mix. Add speaker IDs where the face is missing or two voices share the audio. A name super is a separate overlay, held briefly, not a substitute for per-cue IDs.

Can I use DYNAMIC karaoke on a definition clip?

Only if the mute pass is readable. Definitions run long. Two-line BASIC is the usual webinar-clip burn. Save word-timed looks for a short spoken hook you can actually finish.