Drift Fix
Upload a video and a separately-recorded audio file of the same event - a video's built-in mic plus a higher-quality lav, boom or second-camera recording - and get them synced back together. Unlike tools that only apply a single constant delay, this measures and corrects real clock-rate drift too: the slow, progressive misalignment that builds up over a recording because no two devices' clocks run at exactly the same rate.
Questions fréquentes
What's the difference between "offset" and "drift"?
An offset is a constant delay - the audio file simply started a fixed number of seconds before or after the video, and stays that way throughout. Drift is different: it's a progressive misalignment that grows (or shrinks) over time because the two recording devices' internal clocks run at very slightly different rates. A video might be perfectly in sync at the start and noticeably out of sync a few minutes later - that's drift, and a single constant delay can't fix it.
How does this tool detect drift, not just offset?
It cross-correlates a window of audio near the start of both files and another window near the end, finding the best-matching alignment offset at each point. If those two offsets are the same, it's just a constant delay. If they differ, the difference reveals the clock-rate drift, which is then corrected by very slightly stretching or compressing the separately-recorded audio (a sub-1% adjustment for realistic device clocks) before also applying the constant offset.
What if my two files are only offset, with no real drift?
That's handled automatically and is actually the more common case: if the measured drift is small enough to be measurement noise rather than a real clock-rate difference, only the constant offset is corrected - no unnecessary stretching is applied.
Does this handle a single file with a variable frame rate, or footage stitched from multiple clips?
No - this tool is scoped specifically to the well-defined two-file case: one video and one separately-recorded audio file of the same continuous event. Variable-frame-rate correction within a single file, or aligning many separate clips, is a different and much less well-defined problem that this tool doesn't attempt.
How long do my files need to be?
At least 12 seconds each. Drift is measured by comparing an alignment window near the start of the material to one near the end, so there needs to be enough length between those two windows for the measurement to be reliable - very short clips don't give it enough to work with.
Do the video and audio need to be the exact same length?
No. They just need to overlap in content - the same real-world event, even if one recording started a little earlier or later, or runs a little longer, than the other.
Is my video or audio uploaded to a server?
No. Audio track extraction, the drift analysis, the correction, and the final muxing all happen locally in your browser - the video engine (ffmpeg.wasm) is the same one Video to GIF and Subtitle Burner already use on this site. Nothing is uploaded anywhere.
What does the final download contain?
Your original video's picture track, untouched, with the separately-recorded audio file corrected for drift and offset and muxed in as the new audio track, exported as an MP4.
What if the tool can't find a confident match between my two files?
It will tell you rather than silently guess. A low-confidence result usually means the video and audio aren't actually recordings of the same event, one of them is too quiet or has too little shared audio content, or the drift is larger than what this tool is designed to handle.