Why Your Audio Drifts Away From the Picture, and How to Fix It

Four shapes an out-of-sync video can take

Pick three moments where sound and picture are easy to compare, such as a clap, a door closing or a clearly spoken word on camera: one near the start, one in the middle and one near the end. Note whether the sound arrives early or late at each point, and roughly by how much. Those three readings sort almost every case into one of four patterns.

Constant offset

The sound is equally early or late everywhere, for example a quarter of a second behind from the first frame to the last

Growing drift

Sync is fine at the start and worsens steadily, ending a second or more out by the final minute

Sudden jump

Everything matches until one point, then stays off by a fixed amount afterwards

Device-specific delay

The file looks fine on a laptop speaker but late on wireless headphones, a TV or a soundbar

People tend to notice sound that arrives before the picture sooner than sound that arrives slightly after it, probably because in everyday life light reaches us before sound does. Broadcast engineering guidance reflects that asymmetry by tolerating a little more lag than lead, so a small early-audio error can feel worse than a larger late one.

What causes each pattern

A constant offset usually comes from one step that added or removed time at the start: audio from a separate recorder lined up by eye, a clip nudged on an editing timeline, an export that padded the audio, or a playback chain that adds latency.

Growing drift almost always means two clocks disagree:

• Variable frame rate recordings. Phones and screen recorders often vary the frame rate to save power or keep up with the screen. Software that assumes a constant rate places frames at the wrong times and the error accumulates; the explainer on variable frame rate covers the mechanics.
• Separate recorders. A camera and an external audio recorder each run on their own internal clock, and over a long take the two can slip apart audibly.
• Frame rate mismatches. A 29.97 frames per second clip treated as 30, or 25 fps footage conformed to 24, plays at a slightly different speed from its sound. The 29.97 versus 30 case alone adds about 3.6 seconds of drift per hour.
• Sample rate mislabels. Audio recorded at 48 kHz but treated as 44.1 kHz, or the reverse, plays at the wrong speed. This one also shifts pitch, so it is usually obvious.

A sudden jump points to an edit or a recording fault: frames dropped while recording, two files joined at that point, or a section removed from one track but not the other.

Device-specific delay comes from playback rather than the file. Bluetooth headphones and speakers buffer audio before playing it, and TVs and soundbars can add processing time. Many apps and operating systems compensate for this, but not every combination of device, codec and app does.

Diagnose in this order

1. Play the same file in a second player, such as VLC, on a different device, using the built-in speaker or wired headphones. If the problem disappears, the file is fine and the delay is in playback.
2. Open the original recording rather than the edited export. If the original is in sync and the export is not, the edit or the export settings introduced the error.
3. Measure at the start, middle and end, and decide which of the four patterns you have.
4. Inspect the file in a media inspector such as MediaInfo. Note the frame rate mode (constant or variable), the frame rate and the audio sample rate.
5. If you edited the video, compare the project or sequence settings with the source: frame rate, sample rate and whether the editor converted variable frame rate footage on import.
6. Change one thing at a time, starting with the cheapest fix that matches the pattern, and re-measure after each change.

From a terminal, ffprobe, which ships with ffmpeg, prints the same details. If the two frame rate values it reports differ noticeably, the file is probably variable frame rate:

ffprobe -hide_banner -show_entries stream=codec_type,r_frame_rate,avg_frame_rate,sample_rate recording.mp4

Fixes matched to each cause

Playback delay. Confirm with wired headphones first, then use the player's audio delay control (in desktop VLC it sits under Tools, Track Synchronization) or the TV's lip-sync or AV sync setting. Re-exporting a file to compensate for one pair of headphones makes it wrong for everyone else.

Constant offset in the file. Slide the audio track by the measured amount in your editor, or shift it without re-encoding. This command delays the audio by 0.3 seconds while copying both streams unchanged:

ffmpeg -i input.mp4 -itsoffset 0.3 -i input.mp4 -map 0:v -map 1:a -c copy shifted.mp4

To move the audio earlier instead, put the -itsoffset option before the first input and keep the mapping the same, which delays the picture.

Variable frame rate drift. Convert the footage to a constant frame rate before editing, using the editor's own conversion, a converter such as HandBrake with its constant frame rate option, or ffmpeg. This re-encodes the picture, so keep the original:

ffmpeg -i phone.mp4 -vf fps=30 -c:v libx264 -crf 18 -c:a copy phone-cfr.mp4

Separate-recorder drift. Line up a clap or sharp sound at the start, find the same sound at the end, and stretch the external audio very slightly until both line up. Most editors have a rate-stretch tool for exactly this.

Frame rate mismatch. Set the timeline to the source frame rate and export again. Sudden jumps are fixed by splitting at the jump and re-aligning the later section; if frames were dropped during recording, they cannot be recovered, only worked around section by section.

A worked example: a workshop recording that drifts

Note the order: the sync was repaired in the edit project, not by patching the export. Every later version of the video inherits the fix.

Checking a translated or dubbed export

If the source is in sync, a translated video made from it should keep the picture's timing, because the picture is copied rather than re-encoded. A few things look like sync problems but are not:

• Dubbed lines are fitted to the original speech segments, so a line may start slightly early, run slightly late or be sped up gently, by default up to about 1.15 times. The article on syncing translated audio with video explains that timing fit.
• There is no lip-sync. In close-ups, mouths will not match the translated words, which is a property of voice replacement, not drift.
• Subtitle timing is a separate topic with its own patterns; see subtitle timestamp alignment.

If the whole dubbed video drifts, compare it against the exact file you translated. Usually either the source already drifted, or the comparison is against a different cut.

Pitfalls and limits of sync repairs

• Converting variable to constant frame rate re-encodes the picture, can enlarge the file and duplicates or drops frames to hit the new rate.
• Fixing an export instead of the project means the error returns the next time you export.
• Stretching audio by more than a tiny amount changes how it sounds; large corrections need a different approach.
• Frames dropped during recording are gone. You can re-align sections, not restore them.
• A player's audio delay setting only changes your own playback, never the file your viewers receive.
• Wireless latency differs between headphones, codecs and devices, so a delay value that suits one setup is wrong on another.

What mydubly supports and produces

mydubly accepts MP4, MOV, WebM, MKV and M4V video up to 2 hours per file. The browser decodes the audio locally and uploads only compressed audio chunks; the video itself never leaves your device. From that you can get a transcript, SRT and VTT subtitles, a translated transcript, or a dub: the translated voice over the original music and effects, as an M4A file plus a translated video.

In the translated video, the browser swaps the audio track and copies the picture without re-encoding it, so frame timing stays as recorded, as described in browser video muxing. The new audio track carries the translated voice over the original music and effects, with the original speech removed. If the video codec cannot go into MP4 the output is MKV, and on devices with too little memory for a very large file you receive the translated audio to combine yourself. mydubly has no editing tools, so it cannot repair sync that is already wrong in the source.

When to fix sync in another tool first

Repair sync in your editor before translating, because a dub inherits whatever timing the source has. Screen recordings deserve extra care since they are frequently variable frame rate; the screen recording translation page covers that workflow. The dub already keeps the original music and effects under the translated voice; if you want full control over them, export a dialogue-only version from your project, dub that, and mix the translated voice with your clean music stem in an editor. Keeping background music in translated videos walks through the options and the limits of automatic separation.

Next step: translate a file you have checked

Once the source plays in sync at the start, middle and end, run it through the video translator and check the same three points in the result. For a voice version, AI dubbing costs 50 credits per minute with a 2-minute minimum, so a 10-minute video costs 500 credits (50¢).