How to cut a long session into clips that each end on a spoken cue
Give Vizard Agent a long recording and the spoken cue each clip should end on. Vizard Agent transcribes it, extracts the timing of every occurrence of your cue words, plans cuts that keep the whole session in order with nothing skipped, and renders each clip with captions and any overlay you asked for.
What do you type into Vizard Agent?
State the rules as rules. Vizard Agent treats "chronological", "do not skip any footage" and "end each clip on this word" as hard constraints rather than preferences, so listing them plainly is exactly the right way to brief this job. Say if the first clip needs to be longer for an introduction.
Prompt
What does Vizard Agent actually do?
Some recordings have their own structure built into the speech: a word that marks the end of each section. Cutting on that word gives clips that are self-contained and in order, with nothing dropped between them. Vizard Agent times every occurrence of the words it needs, here "yes", "no" and "thank you", and tests any overlay on a short clip first. In the session behind this article it:
- Pulled the transcript and shot list, read the full transcript with its timestamps and made a lightweight proxy to work against.
- Loaded the word timings, extracted the timing of every occurrence of those three words, and reviewed a contact sheet of the speaker's framing.
- Built the glow overlays and planned the exact clip boundaries.
- Previewed the overlay on a short test clip, retested it with corrected enable times, and compared the two overlay states and a plain frame side by side before accepting the look.
- Generated captions for all ten clips, re-generated them at a safe vertical position, rendered everything and reviewed frames from the middle and end of every clip.
- Delivered ten clips in chronological order covering the whole session, one of them probed at 1600x900, H.264, 30fps and 53.07 seconds.
When does this not work well?
Cue-based cutting is exact, and exactness cuts both ways: if the cue word appears somewhere you didn't intend, the clip ends there. A common word like "yes" in casual conversation produces boundaries you didn't want, and if no cue falls inside the length window, something has to give, so say which.
Common questions
What makes a good cue word? Something you say deliberately at the end of a section. A common conversational word will fire in the wrong places.
Can the first clip be longer? Yes. Say so and Vizard Agent keeps the introduction whole in the opening piece.
How many clips will I get? As many as the cue and the length range produce. Vizard Agent reports the durations when it delivers them.