The quick answer
In Subtitle Edit, use Synchronization → Adjust all times for a constant delay. Choose Show earlier when captions appear late and Show later when they appear early. Use Point sync when different parts need different corrections. Our seconv 5.2.0 check verified a 1.5-second earlier shift in three small SRT files; it did not assess synchronization to speech.
First check whether the delay stays constant
Open the subtitle file alongside the matching video. Compare a clear moment near the beginning, another near the middle and one near the end. Use an audible phrase or visible action with an unmistakable boundary. Do not assume that two files with similar names are the same edit.
If every checked cue is late by about 1.5 seconds, a negative 1.5-second shift is a reasonable repair to test. If the beginning is right but the end is wrong, inspect for drift. A missing introduction, an inserted scene or a different cut can require separate corrections rather than one global adjustment.
- Same error throughout: try a fixed shift.
- Gradually changing error: inspect reference points and use point synchronization.
- Abrupt change after a cut: check that the subtitle file matches the video edit before adjusting a range.
Fix a constant delay without replacing your original
Keep an untouched copy of the original SRT. In Subtitle Edit, open Synchronization → Adjust all times. Enter the amount you measured and choose Show earlier for late captions or Show later for early captions. Check whether the change applies to all lines, selected lines or the selected lines and everything after them.
For the practice files below, choose all lines and shift 1.5 seconds earlier. The first cue should move from 00:00:03,000–00:00:05,500 to 00:00:01,500–00:00:04,000. Its duration should remain 2.5 seconds. Save the result under a new name as SRT with UTF-8 text encoding.
Replay the beginning, middle and end using the output file. Check that the first cue has not been pushed before the start of the video. A mathematically correct shift can still be the wrong repair if the measured delay was wrong.
Use reference points when the error changes
Open Synchronization → Point sync. Select a subtitle line and set its correct position in the video. Add reference points where needed, then apply the change and inspect the intervals between those points. Pick moments you can identify confidently instead of guessing from a quiet or overlapping phrase.
The vendor documents point synchronization for varying drift. We have not tested its desktop controls or interpolation behavior in this observation. If a sequence of captions belongs to footage that is missing entirely, synchronization cannot recreate the missing scene or decide which text should remain.
What our three-file check actually showed
On 3 October 2026 we downloaded the official Windows x64 seconv 5.2.0 release. Its SHA-256 digest matched the digest published for the release asset. We ran the converter on three owned synthetic SRT files with four cues each: English, Simplified Chinese and Spanish.
The requested operation was a negative 1500-millisecond offset with UTF-8 output and a separate output folder. All three commands completed successfully. We independently parsed the actual output files and compared every cue against the expected result: all 24 start/end timestamps moved exactly 1.5 seconds earlier and all 12 cue texts, cue numbers and line breaks were preserved.
These samples test file transformation. They contain no matching video or speech and do not establish transcription accuracy, translation quality, reading comfort or synchronization to real dialogue. The desktop steps above are based on the linked official documentation.
| File | Cues | Observed shift | Text and line breaks |
|---|---|---|---|
| English | 4 | Every start and end: −1.5 seconds | Preserved |
| Simplified Chinese | 4 | Every start and end: −1.5 seconds | Preserved |
| Spanish | 4 | Every start and end: −1.5 seconds | Preserved |
Check the subtitle file in the app that will deliver the video
A corrected SRT is editable text with timing. It is not a rendered video with captions permanently visible. Keep the corrected file if your destination accepts a separate subtitle track. If you need burned-in captions, import it into the editor used for final delivery and inspect that editor’s actual exported video.
CapCut documents SRT import on desktop and web. Its current instructions say the mobile app does not directly import subtitle files. That is a documented platform distinction, not a CapCut test we performed. Our prior CapCut access attempt did not produce an export, so we make no claim here about account entitlements or a successful cross-app round trip.
Inspect the example files
- Download the SRT practice pack
Three original files, three actual seconv outputs, checksums and reproduction notes. No app executable included.
- Read the test result
Version, exact checks, file hashes and the limits of this observation.
What this evidence covers
- The observed test covers seconv 5.2.0 on Windows x64 with 12 synthetic cues.
- No matching audio or video was used, so real dialogue synchronization was not evaluated.
- Desktop controls and point synchronization are documented instructions, not hands-on GUI observations.
- No recognition, translation, burn-in, performance benchmark or successful CapCut export is claimed.
Common questions
Should I move late subtitles earlier or later?
Earlier. If a line appears 1.5 seconds after its corresponding speech, subtract 1.5 seconds from its timestamps, then check the saved result against the video. Move early captions later.
Why are the captions correct at the start but wrong at the end?
A fixed delay may not be the problem. Check for changing drift, a different video edit or missing scenes. Use reference points rather than repeatedly shifting the entire file.
Did Software Signal test automatic captions?
No. We tested a timestamp shift in existing synthetic subtitle text using seconv 5.2.0. We did not test speech recognition, translation, desktop usability or a final captioned video.
Can I use the practice files in my own tests?
Yes. The small practice texts and expected timing exercise were created for this guide. The download includes the original inputs, actual converter outputs, a reproduction command and a sanitized result manifest. The files are examples rather than subtitles for a real video.