The practical answer
Note the source duration, selected range, audio track and output requirements. Inspect a short sample with speech, a pause and a visible scene change. Reopen the actual download, check start and end boundaries, and sample the middle.
Plan the input and the review
- Note the source duration, selected range, audio track and output requirements.
- Inspect a short sample with speech, a pause and a visible scene change.
- Reopen the actual download, check start and end boundaries, and sample the middle.
Choose the operation for this step
- Process an audio file: Choose an available operation and output format, then listen for clipping, missing channels and cut-off speech.
- Process a video file: Check the selected range, aspect ratio and audio presence in a test export; a filename extension alone does not establish playback compatibility.
- Edit a transcript and subtitle timeline: Correct names and numbers against the recording, review cue boundaries and language tracks, then test exported subtitles with the intended player.
- Use the configured server media route: For large recordings, inspect the server-processing notice and service availability before uploading; verify the returned file and avoid assuming the browser-local privacy model applies.
- Inspect media delivery issues: Media analysis uploads to a configured persistent service; subtitle checks run in the browser. Review silence, black frames and loudness in context before making changes.
An editorial example to compare
Make the acceptance condition visible
An automatic transcript labels a product code incorrectly and the last cue extends beyond the trimmed video.
The code is checked against the recording, cue endpoints are aligned to the trimmed duration, and the video with its subtitle track is replayed.
Starting a cut at 12 seconds changes the subtitle origin
In this illustrative example, a recording is trimmed from 00:12.000 to 00:42.000, producing 30 seconds. An original cue runs from 00:15.200 to 00:18.600. Keeping its original timing makes it appear 12 seconds late in the cut. Subtract 12 from both endpoints to obtain 00:03.200 to 00:06.600.
A cue crossing the cut boundary requires a separate decision. An original 00:10.000–00:13.000 cue would become −2 to 1 seconds, so the subtitle workbench rejects that global shift. First remove cues ending before the cut. If the retained speech supports keeping this cue, change its original start to 12 seconds while retaining its original end at 13, then shift the retained cues by −12 seconds: it becomes 0–1. Review and shorten its words accordingly, or remove it if the retained fragment has no useful meaning. At the other boundary, trim original end times to at most 42 seconds before shifting, so the result stays within the 30-second cut.
| Original cue | After subtracting 12s | Review |
|---|---|---|
| 15.200–18.600s | 3.200–6.600s | Check the complete retained speech |
| 10.000–13.000s | −2.000–1.000s | Trim or remove against the actual clip |
| 40.000–44.000s | 28.000–32.000s | Recheck the ending beyond the 30s cut |
A fixed offset only fits an approximately constant error
Compare recognizable speech starts near the beginning, middle and end. If all cues are roughly 0.8 seconds late, try a −0.8-second shift and replay all three positions. If the beginning is 0.2 seconds late and the ending is 2 seconds late, the error is accumulating. One shift will improve one point while worsening another. Check for a different edit, removed pauses or a subtitle file belonging to another media version.
Review recognition text separately from timing. Names, amounts and version numbers need listening checks; cue timing needs checks for starts, ends, overlaps and the next sentence. Accepting an SRT/VTT file establishes its supported format and basic time ranges, not synchronization. Resolve media-version mismatches before adjusting every cue in the wrong track.
Choose a text track or burned-in captions around future edits
A separate SRT or VTT is easy to correct or translate, but the recipient must load the right track. SRT uses commas for milliseconds; VTT uses decimal points and a WEBVTT header. Export the intended format rather than merely renaming the extension. Keep the track with its matching media version and record the corresponding duration.
Burning captions re-encodes the video and places text into its pixels. Supply a font with the needed glyphs; correcting a word later requires another video export. Test a short clip containing the intended scripts, numbers and line breaks for missing glyphs, clipping and sound before processing the full recording. Local transcription, local subtitle burning and translation through /api/translate are separate paths. Confirm translation availability and the text being submitted before using that service.
Check the final result and preserve evidence
W3C specifies WebVTT text tracks and timed cues. Keep timing validation separate from transcript wording checks.
Check moving edges in context when reviewing a visual edit. For animation outputs, inspect frames and crop separately.
Acceptance checks
- Duration, selected range and audio tracks match the requirement.
- Critical speech and subtitle timing were checked against the source.
- The actual final file plays in the recipient’s environment.
Common questions
Does a completed export mean every player can open it?
No. Container, codecs and player support all matter; test the intended environment.
Should every silence warning be removed?
No. Review the recording context and intended pacing before applying a change.
Is speech recognition a final transcript?
Treat it as a draft and verify names, numbers, speaker turns and omissions.
Further reading & sources
Put the guide into practice
Open the relevant tool, preview the result and check your exported file.
