Update subtitles after a video recut

Map the new edit before moving the subtitles

After a video recut, reuse old subtitles only where you can identify the same retained audio in the final video. Make a separate entry for every occurrence of a retained section, calculate its new position, and review cues at the edit boundaries. Removed speech needs removal from the delivery copy; new or replaced speech needs new review. Shifting the whole file cannot handle scenes that move by different amounts.

This guide covers a previously captioned video revised with simple cuts at unchanged playback speed. It is especially useful when reviewed names, terminology or translations are worth preserving. Use media and text you own or are authorized to edit. A single continuous excerpt belongs to the clip-trimming workflow; this task starts when one old track must follow several changed scene positions.

Keep the approved cut and subtitle file together

Preserve the old video, its approved subtitle export and any unsaved project work. Save the final video under a distinct revision name. Record the subtitle language and soundtrack version, too: identical pictures do not prove that narration, dialogue or music remained unchanged.

Ask the editor for a change list, then check it against the actual export. Record each retained interval on the old video's elapsed clock and the start of that occurrence on the final video's clock. Both clocks begin at their respective video starts. If the editor provides frame timecode, convert it using the real timebase before doing arithmetic; frame digits are not milliseconds.

A source-camera timestamp is a third clock. Do not subtract a raw-footage in point from a timestamp in the old finished video's SRT. The mapping needs the clock that the subtitle file actually uses. Keep this small edit map in a separate note or spreadsheet; it is not metadata that an SRT automatically carries.

Give repeated footage a separate occurrence

Consider a fictional craft demonstration. In the old video, scene A occupies seconds 0–10, a discarded explanation occupies 10–20, and scene B occupies 20–30. The new video plays B, then A, then B again as a recap. All retained audio and playback speeds are unchanged in this invented example. These are teaching numbers, not a recording or a Polyglot processing test.

The two B appearances need separate entries even though they use the same old interval. Label them B-first and B-recap. An old-to-new table with only one destination per source scene would silently omit one appearance. Likewise, searching for a repeated sentence cannot tell you which occurrence you are looking at.

Use the final video's order for the map below. The old interval includes its start and stops at its end; each retained section lasts ten seconds. The omitted 10–20 interval has no destination, so its old cues are excluded from the new delivery copy after confirming that the speech is absent.

Occurrence  Old       Final
B-first     20–30 s     0–10 s
A           0–10 s     10–20 s
B-recap     20–30 s    20–30 s

Old 10–20 s: removed
All retained audio: unchanged

Transform both endpoints for each occurrence

For a cue fully inside a retained unchanged-speed section, subtract that section's old start and add its final start. Apply the same calculation to the cue's start and end. The cue duration stays unchanged. Calculate from the preserved old timestamps each time, rather than shifting a result that was already moved for another occurrence.

Suppose old cue 7 is at 22–24 seconds and says “Open the fold.” In B-first it becomes 2–4 seconds; in B-recap it remains 22–24 seconds. Old cue 2 at 2–5 seconds, “Choose a square sheet,” moves to 12–15 seconds in A. There are three final cues from these two selected old cues. The repeated wording is intentional because the matching speech is repeated.

This formula does not cover speed changes, freezes, crossfades or audio that continues independently across picture cuts. A picture edit map can locate the wrong dialogue in those cases. Resolve the actual soundtrack timing, or make a new reviewed draft from the final authorized media, before accepting the affected section.

New time =
  old cue time
  - old section start
  + final occurrence start

Old 7: 22–24 s
B-first: 22–20+0 → 2–4 s
B-recap: 22–20+20 → 22–24 s

Old 2: 2–5 s
A: 2–0+10 → 12–15 s

Hold crossing cues and changed sound for review

A cue crossing a cut is not automatically reusable. For example, an old cue spanning 19–22 seconds crosses B's old start at 20. Applying B-first's offset to the whole cue produces a negative start. Clamping it to zero does not establish which words survive, where the remaining speech begins or whether the sentence still makes sense.

Listen to the final audio around the cut. Keep only the audible wording, establish its real timing, and review whether it needs different segmentation. Do not allocate words by dividing the old cue's duration proportionally. If a cut removes a negation or necessary condition, flag the video edit; silently adding unheard words to the captions is not a repair.

Treat new narration, replacement takes and newly audible transition sounds as unresolved even when the surrounding picture is familiar. W3C's caption guidance includes relevant non-speech audio and speaker identification as well as dialogue. An old sound label may be wrong after the soundtrack changes. A recap might also lose the earlier visual context that identified a speaker, so unchanged words still need a contextual check.

Build the delivery copy in final playback order

For a short file, prepare a separate group of copied cues for each occurrence, resolve its boundary cues, apply its offset once, and assemble the groups in final playback order. Aegisub documents shifting selected rows and both start and end times. Its documentation also warns that negative results become zero, which is another reason to settle boundary cues before shifting them.

Keep occurrence groups separate while applying changes. After shifting B-first, its cues can occupy the same time range that A used in the old file. Selecting the next group by those already-modified timestamps can move the wrong rows. Select by the preserved occurrence membership and check the first and last result of each group.

The Library of Congress describes SRT's numbered blocks, timestamp pairs and text. Give the final blocks unique sequential numbers while keeping an external correspondence note. In this example, old 7/B-first becomes final 1, old 2/A becomes final 2, and old 7/B-recap becomes final 3. Do not remove final 3 as a duplicate merely because its wording matches final 1.

Polyglot can import a prepared SRT for text and timing review and export a separate subtitle file. Its uniform timing adjustment moves project languages together; it is not a way to apply different offsets to separate scenes. The occurrence map and reconstruction described here are manual preparation, not an automatic Polyglot recut, scene-matching or edit-list import feature.

Reuse translations without treating them as automatically approved

Repeat the occurrence mapping for each delivered language using that track's actual timing. A translation with independently edited boundaries can cross a cut that the source cue does not. Preserve its reviewed wording where appropriate, but have a reader of that language review changed sentence context, speaker references and shortened phrases.

Keep two decisions separate: the text may remain approved while its new timing is pending, or its timing may be clear while the wording needs revision. Record those states by language and occurrence. Do not replace a finished translation wholesale just because another section changed, and do not treat reuse of an old approved sentence as approval of the whole new track.

Accept every occurrence against the final export

Reopen the exported subtitle file with the exact final video and reset any player subtitle-delay compensation. Check an identifiable spoken anchor near both ends of every reused section, then play across each cut at normal speed. For repeated scenes, verify each appearance independently. Matching the opening of B-first says nothing about B-recap.

Confirm that removed speech is absent, added audio has been reviewed, crossing phrases remain faithful, and captions neither linger into the wrong scene nor disappear too soon. Validate positive durations, chronological starts, intentional overlaps and the expected cue total. Here the three-cue total is explained by the occurrence map; matching the old count would actually be suspicious.

Deliver one matching file per requested language with the final video revision and the occurrence note. Keep unresolved sections pending. Verify the actual receiving player or upload with its correct language and track selection before marking destination acceptance complete. The final result should explain both what survived from the reviewed old subtitles and what was newly checked for this particular cut.

Keep exploring.