18 September 2026

Fix Audio Out of Sync: Line Sound Up to Picture (2026 Guide)

Your phone writes video and sound on two separate clocks. At 30 frames per second the camera stamps a frame every 33 milliseconds, while the microphone timestamps its own samples on a different counter. Nothing forces those counters to agree, and when they slide apart by half a second, the door slams after it has already closed.

That is the whole problem, and it is why you can fix audio out of sync without a specialist app. The repair is arithmetic: find how far the sound sits from the picture, then either move the sound or re-time it. Two failure patterns cover almost every case, and they need different fixes.

What follows is how to tell them apart in under a minute, the exact commands for each, and the free editor route if you would rather drag a clip with a mouse.

Fix Audio Out of Sync: Constant Offset or Drift?

There are only two patterns. A constant offset means the sound is early or late by the same amount at the start, the middle and the end, so shifting the whole track once finishes the job. Drift means the gap grows as the clip plays: fine at ten seconds, half a second out by ten minutes. Drift needs re-timing, not a shift.

What you see	Constant offset	Drift
First 10 seconds	Sound ahead of or behind the picture by a fixed gap	Sound ahead of or behind by a small gap
Last 10 seconds	The same gap, unchanged	A clearly larger gap
Usual source	A separate microphone, an editing slip, a recorder that buffered, an app that rebuilt the timeline on export	Phone screen recording, a recorder with variable frame rate switched on, some phone cameras
Fix	Method 1 or Method 2: shift the audio track	Method 3: re-time the file at a constant frame rate

Run that test before you install anything. Play the opening ten seconds of the clip, then jump to the end and play ten more. If the error is identical in both places it is an offset. If the end is worse, skip the offset sections and go straight to Method 3.

Measure the Offset in Twenty Seconds

You need one sharp event: a hand clap, a door closing, a light switch, a pen tapped on a desk. Find it in the clip, or film a second of it before the real take. A sharp sound matters because it has one frame you can point at, while a soft one smears across several.

Then open the file in something that shows a waveform beside the picture. CapCut, DaVinci Resolve and iMovie all do, and Audacity opens a video file and draws its waveform without converting it first.

1. Scrub to the loudest, sharpest sound in the first thirty seconds of the clip and note the timecode.
2. Look at the waveform and find the start of the spike that sound made. Note that timecode too.
3. Subtract one from the other. That difference is your offset.
4. Note the sign: sound arriving before the picture, or after it.

At 30 frames per second one frame is 33 milliseconds, so measuring to the nearest frame is as precise as your eye can go at normal zoom. Do not chase the twenty-odd milliseconds of near-silence that some AAC encoders place at the head of a track; that is a codec artefact, not your mistake.

If the offset you measure near the end is larger than the one near the start, no single shift will hold across the file. That is the drift case again, and Method 3 is the only one of the three that applies.

Method 1: Shift the Whole Track with FFmpeg

FFmpeg is free, runs on Windows, macOS and Linux, and does this without re-encoding your video. The command changes the audio timestamps and copies every video frame instead of rebuilding it.

The direction rule is short: in FFmpeg a positive input offset delays the streams of that input, as the official FFmpeg documentation states directly. So a positive offset placed before the audio input pushes the sound later.

If the sound arrives too early

ffmpeg -i input.mp4 -itsoffset 0.4 -i input.mp4 -map 0:v -map 1:a -c copy output.mp4

The same file is fed in twice: video from the first copy, audio from the second. The -c copy flag means neither stream is re-encoded.

I tested this on a three second clip with a known gap. The audio start time moved from zero to 0.476 seconds while the video stayed at zero, and the stray 24 milliseconds is the encoder delay that AAC adds to the head of a track. That sits well below the point anyone notices.

If the sound arrives too late

Cut the head of the audio instead. A trim of the first fraction of a second slides everything that follows earlier:

ffmpeg -i input.mp4 -af "atrim=start=0.4,asetpts=PTS-STARTPTS" -c:v copy -c:a aac output.mp4

The atrim and asetpts pair, both listed in the FFmpeg filter reference, drops the first 0.4 seconds of sound and restarts the clock at zero. On the same test clip the audio track went from 3.000 seconds to 2.518, exactly the amount requested, and the video was copied untouched. The price is one audio re-encode, which is cheap.

If you would rather not touch timestamps at all, insert silence at the head of the track with -af adelay=400:all=1 and keep -c:v copy. That lengthened the audio in my test from 3.000 seconds to 3.518. It solves the early case only, and it re-encodes the audio, so the remux above stays the cheaper option when the sound is ahead of the picture.

Then play the file and check the clap again. If the sound is still off, halve the number and run it once more; the offset is additive, so two passes converge quickly. For the technically curious, ffprobe -v error -show_entries stream=codec_type,start_time,duration -of csv=p=0 output.mp4 prints both streams and shows the difference the command created.

Method 2: Nudge the Audio in a Free Editor

If you would rather see the waveform move, an editor does the same job with a mouse. The steps are close enough to identical in CapCut, DaVinci Resolve, iMovie and Premiere that the labels are the only real difference.

1. Drop the clip on the timeline and scrub to your sharp sound.
2. Detach the audio from the video: right-click the clip and pick the separate or detach audio option.
3. Nudge the audio clip left or right with the frame keys rather than the mouse. One frame at 30 fps is 33 milliseconds.
4. Loop the same two seconds and step a frame at a time until the sound spike lands on the exact frame where the clap happens.
5. Export. Most editors re-encode both streams, so export at a high quality setting.

This is the better route when you also want to trim, crop or caption the same clip in one pass. It is the wrong route for drift, because nudging produces a constant shift and a constant shift cannot follow a gap that grows. Resolve’s free version has plenty of timeline control for this, and CapCut runs on phones.

One practical warning about editors: many timelines refuse to place an audio clip before time zero. If the sound needs to start before the picture does, pad the front of the video with a second of black first, nudge into the padding, and trim the extra second afterwards.

Method 3: When the Gap Grows

Drift usually has one cause: a variable frame rate. Screen recorders and some phone cameras write a frame only when the picture changes, then stamp the file with an average frame rate. A player that believes that average runs the picture at slightly the wrong speed, and the error accumulates, which is why the first minute looks perfect and the tenth does not.

The repair is to give the file a genuinely constant frame rate and let the audio resampler follow it:

ffmpeg -i input.mp4 -fps_mode cfr -af aresample=async=1000 -c:v libx264 -crf 18 -c:a aac output.mp4

fps_mode cfr rebuilds the video timeline at a constant rate. aresample=async=1000 lets the audio be stretched or squeezed by up to 1000 samples per second to stay with it, which is the example FFmpeg itself gives for that option.

On my test clip the output came back with both streams starting at zero and durations of 3.000 and 3.018 seconds. Try it on a copy, because a resampler does something irreversible to the sound: it changes its length, very slightly.

If the gap is large, a second or more across a long recording, splitting beats resampling. Cut the clip at the worst point, correct each half with its own constant offset, then place the two pieces back together on one timeline; joining clips into one file covers that step. Two offsets across one file usually beat one offset that is wrong at both ends.

Long screen recordings are worth re-making rather than repairing. Most recorders have a variable frame rate option that can be switched off in settings, and a clean second take costs less time than guessing at a resample. The screen recording guide goes through those settings per platform.

What Sync-Fix Tools Actually Do, and What They Cannot

Every tool in this space does one of three things: shift the whole audio track by a fixed amount, re-time the timeline, or measure an offset automatically and then do one of the first two. Free editors and command line tools cover all three.

Web pages that promise a one-click repair almost always do the first thing, because a shift is the only fix that can be applied without understanding what is in the file. Automatic measurement is the clever part: a multicam editor, or an app that syncs two recordings from one shoot, finds the same sound in both tracks by comparing waveforms and lines them up.

The limits are worth stating plainly. A shift cannot fix a gap that grows, automatic alignment cannot work without a shared event, and no tool can recover audio that was never recorded. If the microphone was nowhere near the subject, the sound is not late, it is wrong.

One further point is about privacy rather than engineering. Uploading a client’s unreleased footage to a web repair page is a decision to distribute it to a stranger’s server, not just a technical choice. Local tools keep the file local.

Rights and Responsible Use

A video belongs to the person who made it, not to whoever happens to be holding the file. If the footage is yours, everything on this page is ordinary editing. If someone sent you a clip to repair, you have been handed a technical job and not an interest in their work.

Repairing a file is not the same as owning it. Re-uploading a re-synced copy to your own channel is republishing someone else’s work, whoever corrected the timestamps. Keeping a copy for your own reference, fixing a client’s export under their instruction, and redistributing a clip for money are three different situations, and the last two need permission or a licence.

Sync work carries one particular temptation: a repair pass is a convenient excuse to strip a watermark or a credit line. Do not do it. Removing a rights notice is not part of fixing timing, and it turns a routine favour into a violation of the terms you publish under, including YouTube’s terms of service.

One rule covers nearly every case. If you did not record the footage and you do not have permission to use it, treat it as someone else’s work and repair it only for the person who owns it.

Troubleshooting: When the Fix Does Not Hold

The offset is right at the start and wrong at the end

That is drift, and a constant shift can only be correct at one point in the timeline. Return to Method 3 and re-time the file, or split it and correct each half separately.

The file carries more than one audio track

Cameras with two microphones, screen recordings that capture system audio alongside a microphone, and files with a language track all carry several audio streams. The commands above move the stream you map, and -map 1:a takes the first audio stream of the second input, which is not necessarily the one you heard in the editor. List them, then choose by index.

The preview is right but the export is not

Suspect the sample rate. A 44.1 kHz track sitting on a 48 kHz timeline plays at the wrong speed, and a speed error sounds like slow drift rather than a fixed offset. Set the export to 48 kHz and check the result again.

The sound comes from two sources

If the camera recorded its own audio and a separate recorder captured the room, only one of those tracks can line up with the picture. Mute the weaker one and keep the clean recording. Nudging both means fixing one and unfixing the other, because they were never the same signal.

There is no shared event to align to

Narration recorded afterwards, against a silent take, has no reference point. Line up the first syllable by eye and it will still slide out over a long take, because the two devices were never on one clock. Re-record the audio against the picture, or accept the offset.

The audio needs to start before the video

Most timelines will not hold a clip at a negative time. Pad the head of the video with a second of black, nudge the audio into that padding, then trim the padding off before you export.

Frequently Asked Questions

Can I fix audio out of sync without re-encoding the video?

Yes, when the sound arrives early. FFmpeg remuxes the file, moves the audio timestamps and copies the video stream rather than rebuilding it. When the sound arrives late, the audio has to be re-encoded once, because the trim happens on the audio samples themselves.

How much offset do people actually notice?

Roughly 45 milliseconds of sound ahead of the picture, and about 125 milliseconds of lag. Broadcast practice works to tighter limits: the ATSC’s IS-191 recommendation allows no more than 15 ms of lead and 45 ms of lag for television, while the ITU’s controlled tests put detectability at 45 ms lead to 125 ms lag.

Film practice is stricter again, at around 22 ms in either direction. The useful part is the asymmetry: early sound is noticed sooner than late sound, so when you are close, leave the audio a few milliseconds late rather than early.

Does re-encoding a video fix the sync?

Re-encoding a file does nothing at all to its sync. It copies the timeline exactly as it stands, offset included. Only a change to the timestamps moves the sound against the picture, which is why a plain conversion of a broken file comes out just as broken.

Why does my screen recording drift while camera footage does not?

Because screen recorders usually capture at a variable frame rate. The recorder writes frames only when something on screen changes, so the file’s real frame timing is uneven even though the container reports an average. Camera footage is normally written at a constant rate, which is why it keeps time.

Does YouTube Studio have an audio offset tool?

YouTube Studio has no audio offset control in its editor. It can trim, blur and add music, but there is no way to nudge the audio against the picture there. The correction has to happen on the file before the upload, in FFmpeg or in an editor.

Can I do this on a phone?

Yes, if your editing app lets you move an audio clip independently of the video. Separate the audio, zoom the timeline until a single frame is a few millimetres wide, and drag in small steps. Frame-accurate nudging is harder on a touchscreen than with arrow keys, so expect a little trial and error.

What if the two recordings share no sound at all?

Then there is nothing to align them with. Alignment needs a shared event in both recordings, whether that is a clap, a door or the room tone of the same space. Without one, the honest fix is to re-record the audio against the picture.

Which Method Should You Use to Fix Audio Out of Sync?

Sound early by a fixed amount: the Method 1 remux, which is one command and touches no video frame. Sound late by a fixed amount: the atrim trim, or a nudge in an editor if you prefer a mouse. A gap that grows: Method 3 only, and no amount of shifting will substitute for it.

Reach for an editor when the clip needs other work anyway, since the nudge is a side effect of editing you were doing regardless. Reach for the command line when the picture is already correct and only the timing is wrong.

The work itself is not delicate. Measure the gap, move the sound by that number, listen at both ends, and stop when the clap lands on the frame that made it. For a longer set of everyday commands, there are more everyday FFmpeg commands worth keeping nearby.