Captions out of sync: causes and checks

Watch a short stretch near the beginning and another near the end, paying attention to when captions arrive relative to the voice. Most of the time, what you see fits one of three patterns:

A clip trimmed before upload is one known source of a fixed offset. Inside a video file, every frame and every chunk of audio carries a timestamp. Some trims leave those stamps numbered as in the longer source, so the first frame of your clip might read one second rather than zero.

That creates a pair of clocks. By the file's reckoning, the video's first moment is at one second. A browser's video player usually starts at zero on the first frame it displays. Time the words by the file and play them by the player, and every caption shows up a second late, while previewing and in the exported file.

These days Captionwale counts both from the player's zero. Audio goes off for transcription lined up to that point, and the MP4 export places each caption against it too. Older saved projects aren't rewritten, so anything transcribed before the change keeps the timings it had. When such a project is uniformly late, one shift sorts it out.

Go through a clip frame by frame and you may catch a word lighting up just before it is spoken. That is intended. Captionwale draws its captions on a clock set 0.11 seconds in front of the video. A pop or rise entrance needs roughly 0.25 seconds to grow to full scale; started bang on the sound, it would only peak once the word was over. Leading by roughly half the animation's length lands its peak on the sound.

The lead is applied only when captions are drawn on screen, whether in the preview or the exported MP4, and it is identical across styles. Saved timings and SRT or VTT files don't include it. So when judging sync, a caption arriving a fraction early is correct.

Captions also hang on for a moment after their last word, which stops them blinking off. Except on One word styles, a following caption holds back for that moment when the speaker doesn't pause, so in quick speech it may arrive a fraction after its first word has begun.

In the Captions panel, 'Captions early or late?' sits just over the list of captions. It has a separate fix for the first pattern and the second.

To fix a fixed offset, type a number of milliseconds and click 'Shift all'; negative numbers pull captions earlier. All captions move by the figure you entered, your own typed captions too, and nothing can be pushed earlier than the video's first moment. The box goes up and down in 50-millisecond steps, so try a figure, play it back and adjust.

To fix drift that widens, click 'Re-sync to audio'. This reads your video file's soundtrack afresh and realigns the words with the speech, working from where the voice begins and ends rather than from recognising words. Since nothing is transcribed, no minutes are used. The video does have to be on the device you are editing with, and if it isn't, the button doesn't show. Undo it if things get worse.

A few habits take the guessing out of checking:

Shifting only helps an error that is uniform across the video. Applied to drift, it merely relocates the worst of the gap from the beginning to the end, or the reverse. Everything moves together, too, so a title pinned to a certain shot must be re-placed through 'To playhead' or its In and Out times.

Don't expect too much from 'Re-sync to audio'. It targets words given estimated times because a phrase was divided among them, which is how certain transcriptions arrive; right after transcribing, the editor tells you if yours did. Words with timings of their own give it little work, although their edges may still be pulled closer to the sound. Captions you created by hand stay put. And when the voice is hard to make out, under heavy music, whispered or on a poor mic, little or nothing changes.