Audio Sample Rates for Video: 48 kHz, 44.1 kHz and Sync

The short answer

• • 48 kHz for anything with picture: production sound, edit, mix and delivery.
• • 96 kHz when the spec or the sound team asks for it, for example for effects that will be pitched down.
• • 44.1 kHz for music-only releases; resample music to 48 kHz once, before the edit.
• • 48.048 kHz or a 0.1% pull-down only when picture changes speed by 0.1% between shoot and post.

The sample rate itself never causes drift. Drift comes from a speed mismatch between picture and sound, nearly always 23.976 against 24, and it is 3.6 seconds an hour whatever the sample rate.

Why video uses 48 kHz

The Audio Engineering Society's sampling-frequency standard, AES5, makes 48 kHz the recommended rate for professional digital audio and keeps 44.1 kHz for applications tied to consumer products like the CD. The ITU recommends 48 kHz for broadcast sound and for television, SMPTE allows 48 or 96 kHz for television audio, and Netflix wants original-language and M&E mixes at 48 kHz, 24-bit. Both 44.1 and 48 kHz capture everything people hear; 48 kHz wins for picture because it also fits the frame rates.

At 48 kHz every common whole-number frame rate holds a whole number of samples, so a cut on a frame line is also a cut on a sample. At 44.1 kHz, 24 fps gives 1,837.5 samples per frame: editing systems cope, but every edit point has to be rounded.

Samples per frame

29.97 and 59.94 are the awkward ones: 1,601.6 and 800.8 samples per frame. Digital video interfaces deal with it with a five-frame sequence: at 29.97, odd frames carry 1,602 samples and even frames 1,601, so every five frames hold exactly 8,008 (ITU-R BT.1305). It is also why Netflix asks for each segment of a 29.97 or 59.94 IMF package to be a multiple of five frames long: that keeps the audio a whole number of samples.

Pull-down, pull-up and 48.048 kHz

When picture shot at 24 or 30 fps is played at 23.976 or 29.97 (a film transfer, or a true 24 fps camera cut on a 23.976 timeline), it slows by 0.1%. Sound recorded in real time doesn't, so it has to be slowed by the same amount:

Sound Devices' recorders offer both: a 48.048 kHz mode whose files are stamped as 48 kHz, so they play back 0.1% slow, and a 47.952 kHz mode. Pro Tools has matching 0.1% pull-up and pull-down settings. A shoot avoids all of it when the camera records 23.976, the sound recorder runs 23.976 timecode at 48 kHz, and nothing changes speed in post. Pull-down only comes back when the camera really runs at 24.000 (film, or a digital camera set to true 24) and post works at 23.976. Settle that in prep: the camera report, the sound report and the edit timeline should name the same rate, as Netflix's sound spec also asks.

Audio for the 24 → 25 speed-up

Playing 24 fps material at 25 fps speeds the sound up by 4.17% and raises its pitch by about 0.71 semitone. Picture and sound stay in sync as long as both speed up together; the question is whether to correct the pitch, and whether to conform the mix from its stems instead. In Pro Tools, the 4% pull-up at 48 kHz plays the session at 50,000 Hz (48,000 × 25/24). The pulldown and conversion guide has the speed, running time and pitch for each conversion.

Keeping sound in sync in post

• • Sync by timecode. Production recorders write Broadcast WAV (BWF) files whose header stores the start time as a count of samples since midnight; 10:00:00:00 at 48 kHz is 1,728,000,000 samples. The iXML chunk alongside it names the timecode rate (23.976 is written as 24000/1001) and whether it is drop-frame. NLEs read both and line the sound up with the picture timecode. The timecode on set guide covers how the timecode gets there.
• • Check the first and last clap. If a long take is in sync at the slate but not at the tail, it is a speed mismatch, not an offset: look at frame rates before you nudge anything.
• • Resample once. Convert 44.1 kHz music and any odd-rate files to 48 kHz before they go into the edit, rather than letting each system resample on the fly.
• • Turn over at the delivery rate. The AAF or OMF to the mix stage should be 48 kHz at the project frame rate, with the same frame rate on the reference picture.

The Speed Change calculator in CineMath shows the 0.1% directly: 24 to 23.976 is 99.90% speed and 100.10% duration, and it warns when a conversion mixes fractional and whole-number rates. The free speed change calculator works it out in your browser too.

Audio sample rate questions

Next: set the camera and recorder to agree before the first take. The timecode on set guide walks through jam sync and record modes, and the frame rates guide explains which rate to agree on.