Every camera and audio recorder keeps time with its own crystal, and no two run at exactly the same speed. Syncing lines two recordings up at one moment; from there they drift apart at a steady rate, 3.6 milliseconds an hour for every part per million (ppm) between their clocks. Two devices 20 ppm apart end a 45-minute ceremony 54 ms out of step and a four-hour reception file 288 ms out, against a broadcast limit of 40 ms for sound that arrives early. Start a new file for each part of the day, check the sync at the end of every long file as well as the start, and use timecode where you can.
The drift arithmetic in this post is ours. The thresholds come from the European Broadcasting Union and the International Telecommunication Union, and the clock figures from the manufacturers' own pages. None of them can tell you how far apart your own camera and recorder are, so the post ends with a test that measures it.
Why two recordings drift apart
A recorder set to 48 kHz takes 48,000 samples for each second its own clock counts. If that clock runs 20 ppm slow, it counts each second a little late and spreads those 48,000 samples over 1.00002 real seconds. Your editor plays the file back at exactly 48,000 samples a second, so the recording comes out 20 microseconds short for every second of the day. The camera's clock has its own small error, and what matters is the difference between the two.
Syncing does not change that. Final Cut Pro's "Use audio for synchronization" option, in Apple's description, makes precision sync adjustments using audio waveforms, and Premiere Pro and DaVinci Resolve offer the same kind of waveform and timecode sync. Each one finds where two recordings match and lines them up. Lining clips up moves them; it does not change either one's speed, so a gap that grows across the file is still growing after the sync.
The direction matters. If the recorder's clock runs slow against the camera's, its file comes up short and the sound arrives progressively early. If it runs fast, the sound runs progressively late. Viewers notice early sound much sooner, as the standards below show.
How far it drifts
Drift is the clock difference times the length of the file, and one ppm is 3.6 ms an hour. On our arithmetic, with the clock differences as illustrations rather than measurements of any device:
- 5 ppm apart: 18 ms an hour, 14 ms across a 45-minute ceremony, 72 ms across a four-hour reception file.
- 10 ppm: 36 ms an hour, 27 ms across the ceremony, 144 ms across the reception.
- 20 ppm: 72 ms an hour, 54 ms across the ceremony, 288 ms across the reception.
- 50 ppm: 180 ms an hour, 135 ms across the ceremony, 720 ms across the reception.
For scale, one frame at 23.976 frames per second lasts about 41.7 ms, and one at 29.97 about 33.4 ms. A 20 ppm gap puts the end of a 45-minute ceremony more than a frame out, and the last toast of a four-hour reception file about seven frames out.
When viewers notice
EBU Recommendation R37 (2007) sets the end-to-end limits for television at 40 ms with the sound before the picture and 60 ms with the sound after it. The EBU set those values from subjective tests of the delay at which a failure of lip sync becomes perceptible to 50% of observers. ITU-R Recommendation BT.1359 (1998) puts the average detectability thresholds at about 45 ms with the sound early and 125 ms with it late, and the acceptability thresholds at about 90 ms and 185 ms.
Both say the same thing about direction: an audience tolerates late sound for much longer than early sound. So the drift to worry about is the one that makes your recorder's audio run early, and 40 ms is the number to plan around. At 20 ppm, a file reaches it 33 minutes in.
What the spec sheets say
Devices built for timecode publish their clock accuracy. Zoom lists the F6's timecode precision as ±0.2 ppm, held by a temperature-compensated crystal oscillator whether the recorder is on or off, and notes that in its external timecode modes the audio clock can be synchronized to timecode. Tentacle rates the Sync E at less than one frame of drift in 24 hours, from -30°C to +85°C, with a 0.2 ppm oscillator. Zoom says its TCA-1 adapter keeps the F3 within half a frame over 24 hours.
At 0.2 ppm, drift is 0.72 ms an hour, which is nothing on a wedding day. Small recorders without timecode simply do not say: the spec pages for Zoom's F3 and Tascam's DR-10L give no figure for the recorder's own clock, and the F3's page offers the timecode adapter instead. A missing number is not a bad clock. It is a reason to measure.
Timecode answers where each file starts. Whether it also stops drift inside a long file depends on whether the device runs its clock from the timecode, which is what Zoom's note about the F6's audio clock is telling you. A device stamped with jammed timecode but running on its own clock can still drift inside a long take, so the end-of-file check applies either way.
Measure your own gear
- Record an hour with the camera and the recorder side by side, at the settings you use at weddings.
- Clap sharply at the start and again at the end.
- In your editor, sync on the first clap, then measure how far apart the two recordings of the last clap are, in milliseconds.
- Divide that offset by the hours recorded and then by 3.6. A 72 ms gap after one hour is 20 ppm.
- Note which way it went. If the recorder's clap lands before the camera's, its sound runs early.
Then divide 40 by your ppm times 0.06 for the minutes a file can run before it reaches the EBU's 40 ms: about 133 minutes at 5 ppm, 67 at 10, 33 at 20 and 13 at 50. Crystal clocks move with temperature, which is why Tentacle states its figure over a temperature range, so leave margin for a hot afternoon ceremony or a cold evening outside.
Keeping a wedding day in sync
- Start a new file for each part of the day: processional to recessional, each block of toasts, the first dance. Every file you sync starts from zero drift.
- Sync on a sharp sound near the start of each file, then play its last words. If they still match, the sync held.
- If the end has slipped, cut the recorder clip at a pause near the middle and sync the second half again on its own. Each half of an hour-long file drifts half as far.
- On long ceremonies and receptions, run timecode on the devices that accept it, and on a recorder that can run its audio clock from timecode, use that mode.
- Do not expect the recording format to help. 32-bit float protects your levels, not your timing; drift comes from the clock.
Check the long, real-time edits first. A full ceremony or a full set of toasts in a doc edit is where drift shows, because nothing is cut away to hide it. Which recorders to run, and where to put them so the vows exist twice, is in the wedding video audio post.
Where Pixally Prep fits
Pixally Prep does the syncing before you open your editor. For each part of the day it builds a timeline with every camera and audio recorder lined up and stacked, nothing cut, in a project for Premiere Pro, DaVinci Resolve or, on a Mac, Final Cut Pro. The first wedding is free, one per person and one per computer, and each one after that is $8.99. Whatever syncs your footage, the end-of-file check above is how you know it held. If you are still choosing an editor, Resolve Studio against Premiere covers what each costs and decodes.
Common questions
- Why does my audio drift out of sync over a long wedding recording?
- Because the camera and the recorder each keep time with their own clock, and no two run at exactly the same speed. Syncing lines them up at one point, and from there the gap grows at a steady rate: 3.6 milliseconds an hour for every part per million between the clocks. Long files, like a full ceremony or a four-hour reception, show it most.
- How much audio drift is noticeable?
- EBU Recommendation R37 sets 40 ms with the sound early and 60 ms with the sound late as the end-to-end limits for television, based on when half of viewers notice. ITU-R BT.1359 puts average detectability at about 45 ms early and 125 ms late. Early sound is noticed far sooner than late sound.
- Does timecode stop audio drift?
- Timecode fixes where each file starts. It stops drift inside a long file only on a device that runs its clock from the timecode; Zoom says the F6 can synchronize its audio clock to timecode in its external modes. Timecode generators such as the Tentacle Sync E are rated at less than one frame of drift in 24 hours.
- How do I fix audio that drifts out of sync in Premiere Pro or Resolve?
- Sync on a sharp sound at the start, then check the end. If the end has slipped, cut the recorder clip at a pause near the middle and sync the second part again on its own, which halves the drift in each part. For future weddings, start a new file for each part of the day so no single file runs long enough to drift far.
- How long can I record before audio drifts out of sync?
- Divide 40 by your clock difference in parts per million times 0.06 for the minutes until the drift reaches 40 ms: about 67 minutes at 10 ppm and 33 at 20. Measure your own camera and recorder by recording an hour side by side with a clap at each end.
- Does 32-bit float recording prevent sync drift?
- No. 32-bit float is about level: it keeps loud and quiet sounds usable without careful gain settings. Drift comes from the recorder's clock, and the recording format does not change it.
