Audio video sync offset calculator

Enter the detected audio offset in milliseconds using a positive value when sound arrives before the matching picture, or a negative value when sound arrives afterward. The result states whether to delay audio or delay video, gives the non-negative delay to apply, and preserves the signed measurement for reference. This makes a commonly confusing sign decision explicit before you configure an editor, transcoder, player, or muxing workflow.

Describe the detected offset with one consistent sign convention

Begin with a measurement that compares the audio event with the matching visual event. A hand clap, slate strike, door closing, or spoken consonant with a clear mouth movement can provide a useful reference. Enter a positive number when the audio is early: for example, enter 180 when the sound is heard 180 milliseconds before the corresponding frame. Enter a negative number when the audio is late: enter -240 when the picture reaches the reference event 240 milliseconds before its sound. This convention describes the position of audio relative to video, not an option copied from a particular editing application. That distinction matters because applications label offsets differently. Some tools ask when an input should begin, while others ask how much delay to add, and their positive signs may mean opposite things. By recording the detected relationship first, you keep the observation independent of any software interface. The calculator preserves that signed measurement in its result and separately provides the corrective direction and magnitude, so the original evidence and the action cannot be confused.

Apply the correction to the stream that arrives early

Synchronization is restored by delaying the stream that arrives first. When the entered value is positive, audio leads the image, so the result says delay_audio and returns that same magnitude as audio_delay_ms. When the entered value is negative, audio lags behind the image, which means video is early; the result therefore says delay_video and uses the absolute magnitude as video_delay_ms. The offset_ms field is always non-negative and expresses the amount of delay to apply. The other stream receives a delay of zero, making the intended operation explicit for automation. A zero measurement returns none because neither stream needs to move. This calculation does not stretch, resample, trim, or otherwise alter media. It computes a start-time alignment correction only. If synchronization begins correctly but drifts progressively during playback, a fixed offset is not sufficient; investigate mismatched sample rates, frame-rate interpretation, clock drift, or damaged timestamps. For a stable lead or lag, however, applying one delay to the early stream aligns the corresponding events without changing playback speed.

Transfer the result safely into an editor or muxing workflow

Use direction to decide which track or stream receives the delay, then use offset_ms as the magnitude. Before entering it into another tool, confirm whether that tool accepts milliseconds, seconds, frames, or timeline timecode. Convert units without changing the correction described by this result: 250 milliseconds is 0.25 seconds, while a frame conversion also depends on the project frame rate. If an interface uses a signed shift rather than a named delay, consult its documentation and map the named direction deliberately, because its sign convention may differ from the detected_audio_offset_ms convention used here. Test the corrected output around several sharp events near the beginning and end of the program. Multiple checks help distinguish a constant offset from gradual drift and reveal measurements based on an ambiguous sound. Keep the original signed measurement alongside the correction in job metadata or review notes. That record explains why a stream was delayed and makes later revisions reproducible. The calculation is deterministic, requires no media upload, and can be inserted into quality-control scripts that already receive an offset from waveform analysis or manual review.