Caption troubleshooting · 8 min read
Fix Subtitles Out of Sync: Offset, Drift, and Cuts
Diagnose subtitles that appear early, late, or drift over time. Use worked timing examples to repair SRT or VTT captions and verify the final video.
By Zhang at Wordtake. Zhang works on Wordtake and writes practical guides to reviewing transcripts and caption exports. Examples in this guide are illustrative; product references are disclosed.
Identify the timing pattern before changing the file
To fix subtitles that are out of sync, compare the same spoken phrase with its caption near the beginning, middle, and end of the final video. A similar error everywhere calls for a fixed time shift. An error that grows steadily may need a timing scale correction. An abrupt jump after an edit usually needs a section-specific repair. Changing every timestamp before diagnosing the pattern can fix one passage while breaking another.
This guide uses illustrative timing examples, not measured product benchmarks. You will need the exact video you intend to publish, its SRT or VTT caption file, and a caption editor that lets you inspect cue start and end times. Keep an untouched copy of both files. A media player can help diagnose a delay, but a playback setting alone does not repair the file you upload elsewhere.
- Fixed offset: captions remain roughly the same amount early or late throughout.
- Progressive drift: captions begin close to the speech but increasingly diverge.
- Edit discontinuity: timing changes suddenly after a removed or inserted segment.
- Local cue error: one phrase is mistimed while nearby captions remain aligned.
Check that the video and captions share the same edit
Open the actual delivery video, not an earlier editing preview. Compare its duration, opening words, transitions, and closing words with the source used for transcription. An added title card, shortened introduction, removed answer, or inserted sponsor message changes the relationship between source time and final-video time. A matching filename is not proof that the files match.
If the audio is already out of sync with the picture, address that problem first. Captions should follow the intended speech timing; they cannot repair a video whose audio track is misplaced. Similarly, check which subtitle track the player is showing. A stale track or previously configured playback delay can make a corrected file appear wrong.
Create a small anchor log. For each check, write the exact phrase, the caption start time, and when the phrase begins in the video. Use distinctive speech with a clear onset rather than background music or an ambiguous pause. Add an anchor immediately before and after any suspected edit. These observations determine which correction is justified.
Repair a constant offset with one signed shift
For a fixed offset, calculate the shift as the desired video time minus the current caption time. Add that value to both the start and end of every affected cue. If a phrase starts at 12.400 seconds in the video but its caption begins at 10.900 seconds, the shift is +1.500 seconds: the caption needs to appear later.
In this example, a cue running from 10.900 to 13.100 becomes 12.400 to 14.600. Its duration remains 2.200 seconds. If the caption instead starts at 14.000 and the speech starts at 12.400, the shift is −1.600 seconds. Always reason from the desired timestamp: editors and players may label delay controls differently.
Before applying the shift to the whole track, verify that two other anchors need approximately the same correction. Save a new file, reset any player-only delay to zero, and preview it against the final video. Inspect the earliest cues when moving captions earlier. A cue that would start before zero needs individual handling; blindly clipping it to zero can alter its duration or create a collision.
- Formula: corrected time = original time + shift.
- Apply the same shift to cue starts and ends.
- Recheck at least one early, middle, and late phrase after saving.
Use two anchors only when the drift is consistent
A growing mismatch cannot usually be solved with a single offset. If the final video has a different playback speed from the caption source, a linear mapping may help. Choose two widely separated matching phrases and calculate a scale and shift. Then test several independent phrases between them. Two aligned endpoints alone do not prove that everything between them is correct.
For an illustrative example, suppose a phrase at subtitle time 10 seconds belongs at video time 12 seconds, while a phrase at subtitle time 610 belongs at video time 618. The scale is (618 − 12) / (610 − 10) = 1.01. The shift is 12 − (1.01 × 10) = 1.9 seconds. The mapping is therefore corrected time = 1.01 × original time + 1.9. A subtitle timestamp of 310 seconds maps to 315 seconds.
Apply a verified mapping to both ends of each cue and inspect the new durations. Prefer the caption editor’s synchronization controls to manually rewriting a long file. If the middle anchors do not agree with the mapping, stop: an edit, local alignment error, or mismatched source may be involved. Do not pick a frame-rate conversion preset merely because drift exists. First confirm the actual source and export timing.
Repair timing jumps at the edit boundary
When timing changes abruptly, split the diagnosis at that boundary. Imagine an original video loses a five-second passage between source times 120 and 125 seconds. Captions before the cut keep their times. Captions entirely after it move five seconds earlier. Captions inside the removed passage must be removed, and any cue crossing the boundary needs manual review.
A global shift would damage the correctly timed beginning. A global scale would distribute the five-second difference across the whole recording instead of repairing the cut. Keep a list of edit boundaries and their cumulative offsets when several sections have changed. Verify the first phrase after each cut and the last phrase before the next one.
For extensive rearrangement, generating or aligning captions against the final export may be less error-prone than patching an old track. Bring verified spellings and reviewed wording forward, but check them against the new video. Reordered material can change context, speaker attribution, and the wording needed for an intelligible caption.
Separate caption-file repairs from player adjustments
A playback offset is useful for testing whether a constant shift solves the problem. It should not be treated as a portable correction. Save or export the repaired caption track, open it in a fresh player session with no extra delay, and confirm that the file itself contains the intended timestamps.
On YouTube, the official caption editor supports changes to cue start and end times. Open the video under Subtitles in YouTube Studio, edit the relevant language track, and adjust its timing fields; use Edit timings if those fields are hidden. Publish the changes when review is complete. YouTube also documents downloading a caption file, editing it, and uploading the corrected version.
Keep the original caption format unless your destination requires a conversion. Renaming an SRT file to VTT does not convert its structure. Check the destination’s accepted formats, preserve the file encoding, and reopen the export before uploading. A successful upload only confirms that the platform accepted a file; it does not confirm that the captions match the speech.
Verify readability after the timestamps are fixed
Correct timing is necessary, but captions also need readable phrasing and appropriate context. Watch the full final video with the repaired track. Look for cues that flash too briefly, linger into unrelated speech, collide with adjacent captions, or break a phrase awkwardly. Review speaker changes and meaningful non-speech information where it is needed to understand the recording.
Check the final published player as well as the local export. Select the intended language track and test your logged-out viewing experience. Then check every edit boundary and a few phrases distributed through the video. Keep a record of the final video version, caption filename, correction applied, reviewer, and unresolved issues so the next upload does not accidentally restore an older file.
Wordtake can provide a timestamped transcript and SRT or VTT export as a starting point. Review the words against the recording, then use your caption editor or publishing platform for any required timing repairs. Save the checked caption file beside the final video so future clips and revisions start from a known version.
- The final video and caption file describe the same edit.
- Cue start times are nonnegative and each end follows its start.
- Any overlaps are intentional and render correctly in the destination player.
- Opening, middle, closing, and edit-boundary anchors match the speech.
- A fresh playback session works without an extra subtitle-delay setting.