Audio pull-up and pull-down

Audio pull changes the timing of sound.

An audio pull changes playback speed to match a deliberate picture-speed change. It can involve a recorder’s sample clock, a file’s declared sample rate, or processing in post. Those are different operations. A useful instruction states the recording clock, file stamp, playback interpretation and required result, rather than simply “pull down the audio.”

Audio pull is also distinct from 2:3 picture pulldown. A field cadence distributes film pictures across video fields; an audio pull changes the relationship between samples and elapsed time. A transfer can require both, but applying a picture cadence does not by itself define the sound treatment.

For the common fractional-rate relationship, speeding up uses 1001/1000 = 1.001, exactly a 0.1% speed increase. Slowing down uses its inverse, 1000/1001 ≈ 0.999000999, a speed reduction of approximately 0.0999001%. Production shorthand often calls both directions “0.1%”; keep the exact ratio in the technical handoff.

Count samples, then specify their clock.

For uncompressed PCM audio, duration is sample count ÷ samples per second. Here, sample count means the number of sample instants per channel, not the sum of every channel’s values. A stereo recording has two values at each instant; adding a channel does not double its running time.

Sample rate, sample count and bit depth are independent quantities. Rate describes the time grid; count describes how many instants are stored; bit depth describes the representation of each value. Changing 24-bit audio to another bit depth is not an instruction to change duration. A metadata edit can leave the sample count untouched while changing the time assigned to those samples.

Recording at 48,048 Hz captures 1.001 times as many samples per real second as recording at 48,000 Hz. If those samples are subsequently played at 48,000 per second, without sample-rate conversion or pitch correction, the recording becomes longer and lower in pitch. Recording at 48,000 and playing the unchanged samples at 48,048 produces the opposite result: shorter and higher in pitch. These statements assume the intended clock rates are actually achieved.

Two pulls, with the direction made explicit.

These examples retain every sample and change only the rate at which those samples are interpreted. They describe a speed change, not ordinary sample-rate conversion.

In the first case, the duration increases by exactly 0.1%. In the second, duration decreases by approximately 0.0999001%. Percentage duration changes are reciprocals of speed changes, not the same percentage with a reversed sign.

For a longer check, 172,972,800 samples recorded at 48,048 Hz represent exactly 3,600 seconds. Interpreting them at 48,000 gives 3,603.6 seconds: 3.6 seconds longer. Use sample counts to avoid accumulating rounded millisecond estimates. Real edits and exports end on integer sample boundaries; a theoretical duration can require rounding by a fraction of one sample. One sample at 48 kHz lasts approximately 20.833 microseconds.

A new header is not new audio samples.

A file stamped as 48 kHz tells a reader how to interpret its stored sample sequence. Changing that stamp, or overriding its interpretation, does not itself calculate replacement samples. It changes duration and pitch when the reader uses that new rate directly.

Ordinary sample-rate conversion, or SRC, instead calculates samples on a new time grid to preserve the correctly interpreted sound’s duration and pitch, within conversion and boundary-rounding limits. Converting a correctly identified, ten-second 48,048 Hz recording to 48,000 Hz therefore produces approximately 480,000 samples per channel and ten seconds of sound. It does not produce the 10.01-second pull in the table. Reinterpreting first and converting afterward can bake the changed timing into another file; operation order matters.

Adobe Audition’s sample-type documentation distinguishes its Interpret Sample Rate and Convert Sample Type commands. Their names are a useful reminder to establish whether a command changes interpretation or processes the samples. Import, conform and export behaviour must be checked in the actual application and version.

Recorder flags need equally specific reading. The Sound Devices 744T manual, firmware revision 2.67, distinguishes 48.048 kHz from its 48.048kF compatibility mode: the latter records at 48,048 Hz but stamps the file at 48 kHz. That mode also has particular timecode restrictions and stamps. Do not carry its settings into another recorder by name alone; confirm the model, firmware, metadata and post-production agreement.

Match the sound decision to the picture decision.

If 24 fps pictures are played at 24000/1001 fps with every picture retained, their duration increases by 1.001. Sound that originally matched those pictures needs the same duration change if it is to follow that new motion. Sound Devices’ file-format overview describes 48.048 kHz recording and 48 kHz playback for this type of workflow.

That does not mean every 23.976 project needs pulled audio. A scene acquired and played at the same fractional picture rate can use sound running at normal speed. Establish what actually changed between acquisition and the intended master. A wrong start offset and a progressive duration mismatch need different remedies.

A 24-to-25 frame-preserving speed-up is larger: speed becomes 25/24, approximately 4.1667% faster, and duration becomes 24/25, exactly 4% shorter. Do not substitute a 0.1% pull. Decide separately whether to retain the resulting pitch change or use pitch-preserving time processing.

Preserve original files and record the chosen clock rate, declared rate, timecode rate/counting mode, ratio, pitch policy and processing order. Test a head sync event and a later sync event after the actual import and export. A correct first slate alone cannot prove that the entire take runs at the right speed.

Calculate the plan before processing media.

The conform calculator helps compare picture durations when a fixed frame count is played at another rate. Use that result to state the required sound-duration ratio, then implement and verify the audio treatment in the recorder or sound application.

Timecode Pro is a calculator; it cannot process audio, resample a WAV, change a recorder’s clock or rewrite production files. Its format calculations do not verify a media file’s metadata or an NLE’s import behaviour. Keep the physical sample-rate equations above with the sound report.

For the separate jobs of labels and clocks, read timecode, genlock and word clock. For the picture-transfer context, continue with film footage and digital playback.