Review timed captions for a short training clip
Fix the source and the review question
A caption file can contain the right sentences while showing them during the wrong instruction. Review the text as it appears with the media, not only as a readable transcript. Keep the source version and the caption version together so a correction can be traced to the sequence it concerns.
Consider a fictional five-second training-clip plan. From 00:00 to 00:02, Speaker A says Choose card B. From 00:02 to 00:03, two knocks sound. The exercise brief defines those knocks as a retry signal. From 00:03 to 00:05, Speaker B says Use card A instead. These are invented script intervals, not a recorded clip or a completed playback test.
The planned review asks whether a viewer receives those instructions, the speaker change and the relevant sound at their intended moments. It does not assess a real trainee or report learning results.
Connect each audio event to its caption cue
W3C WAI’s Captions/Subtitles guidance describes captions as synchronized text for speech and non-speech audio needed to understand media. Its Captions (Prerecorded) explanation also includes identifying speakers and meaningful sounds. Those principles inform this review, without establishing an accessibility certification.
Create a small event-to-cue record in an ordinary document. For each event, retain the source interval, spoken words or sound, relevant speaker identity, current caption interval and finding. Keep uncertainty explicit if a word or speaker cannot be established from the permitted source.
In the deliberately faulty draft, Choose card B appears from 00:02 to 00:04. Use card A instead appears from 00:04 to 00:06 and is attributed to Speaker A. No cue represents the knocks. The speech text alone can look complete while the sequence remains wrong.
Compare beginnings, endings and speaker changes
For an actual review, play the same source and caption versions in the intended player. Inspect when each cue appears and disappears relative to the corresponding audio. Pause or replay a transition when needed, then watch it in motion again. Script timestamps alone cannot establish a recorded clip’s timing.
In the fictional plan, the first draft cue begins after Speaker A has finished. The second extends beyond the planned five-second clip and names the wrong speaker. Record those as separate timing and attribution findings rather than one vague comment that captions feel late.
Check whether the two knocks need representation to preserve this exercise’s retry signal. Describe the sound without inventing words spoken during it. Captions do not, by themselves, supply every visual-only instruction; note that separate gap if the picture carries information the audio never communicates.
Repair a cue and replay its neighbors
Propose matching the first cue to 00:00–00:02, a two-knocks cue to 00:02–00:03, and Speaker B’s instruction to 00:03–00:05. These are corrections to the declared toy plan, not claims about actual media rendering or a universal caption-duration rule.
After editing a real cue, replay the preceding and following events as well. A changed end can cover the next instruction; a moved cue can leave its speaker attribution attached to the wrong turn. Check that the displayed captions do not hide information needed for the task.
Leave a bounded repair record
Retain the reviewed versions, event-to-cue findings, proposed changes and remaining uncertainty. A real finish check includes watching the complete short sequence after repairs, not merely rereading the edited text. Leave unresolved audio or display behavior with a named reviewer rather than guessing.
Use approved existing media tools for such work. This practice assumes no Kiguri caption generation, upload, editing, recording or playback feature, and no completed caption review is represented by the fictional plan.
