Why subtitles drift out of sync

Captions that are right at the start and wrong at the end are a different problem from captions that are wrong everywhere by the same amount. The size and shape of the error tells you the cause, and only one of the two is a transcription problem.

First: constant offset, or growing drift?

Check a caption near the start and one near the end. If both are late by the same amount, you have a constant offset. If the one at the end is much further out than the one at the start, you have drift. They have different causes and different fixes.

Growing drift is a frame rate mismatch

Drift proportional to elapsed time means the captions were built for one frame rate and played at another. The usual pair is 23.976 against 24, or 29.97 against 30: a 0.1% difference, which is one frame every 33 seconds and about 3.6 seconds per hour.

A larger ratio means a bigger mismatch. 25 against 23.976 is about 4%, which is two and a half minutes of error per hour and obvious within the first minute.

• Roughly 0.1% out, about a second every 17 minutes: an NTSC pair, 23.976 against 24 or 29.97 against 30.
• Roughly 4% out: 25 against 24, a PAL and film mismatch.
• Roughly 4.3% out: 25 against 23.976.

The fix is to re-export the subtitles at the frame rate of the sequence, not of the source video. They are not always the same: a clip is conformed when you drop it on a timeline of a different rate, and your captions are not.

Constant offset is a starting point problem

If everything is late by the same two seconds, nothing is wrong with the timing model. Something moved the zero point: the subtitle track was placed at the wrong position on the timeline, the video has a slate or countdown the transcript did not include, or the sequence starts at a timecode other than zero.

In Resolve and Premiere, this is fixed by moving the subtitle track rather than re-exporting. Exports here start at zero, so a sequence starting at 01:00:00:00 will need the track offset to match.

Captions that are wrong in the middle only

Neither of the above. Individual lines landing wrong while their neighbours are correct is a transcription and segmentation problem: a line split at a point the timing does not support. This is what per-word timing prevents: with a time for every word, splitting a line gives both halves times the audio actually supports rather than an estimate from character counts.

Frequently asked questions