What Sample Rate Should My Audio Be?

Every export dialog asks. Most people pick whatever was there last time, and most of the time that is fine — but the two cases where it is not are worth knowing, because neither announces itself.

What the number actually caps

A sample rate sets one hard ceiling: the highest frequency the file can carry is half the rate. That half is called the Nyquist limit, and nothing in the recording chain can put content above it.

44.1 kHz carries up to 22.05 kHz
48 kHz carries up to 24 kHz
22.05 kHz carries up to 11 kHz
8 kHz — telephone — carries up to 4 kHz

Human hearing tops out somewhere around 20 kHz in youth and falls from there, so 44.1 and 48 both have headroom over anything a listener can hear. That is why both are standard and why the choice between them is usually about what the rest of the pipeline expects: 48 kHz for video, 44.1 kHz for music distribution.

Why 96 kHz rarely earns its size

Doubling the rate doubles the file. What it buys at the listening end is content between 24 and 48 kHz, which no one can hear.

It is genuinely useful while you are working — pitch-shifting, time-stretching and some nonlinear processing behave better with the extra room, and a recording you will heavily manipulate is worth capturing high. It is almost never useful in a delivery. Exporting a finished master at 96 kHz doubles every listener’s download to carry frequencies their speakers cannot reproduce.

Telling a rate problem from a bitrate problem

Here is the case that catches people out. A file is reported as stopping at 9.3 kHz. Is that a bad export?

It depends entirely on the rate. If the file is 22.05 kHz, its ceiling is 11 kHz — so 9.3 kHz is 85% of everything it could possibly carry, and the file is doing fine. A spoken-word recording at that rate is a perfectly ordinary thing.

If the same 9.3 kHz came from a 44.1 kHz file, the ceiling was 22.05 kHz and the content stops at less than half of it. That is a low-bitrate stage somewhere in the history, and no re-export will recover it.

This is why the frequency range is judged against the file’s own Nyquist limit rather than against an absolute number. An 8 kHz telephone recording is not a broken file; it is a telephone recording.

What to actually pick

For a delivery: 48 kHz for anything with picture, 44.1 kHz for music, and let the platform resample if it wants to. For a capture you intend to process hard: higher is defensible. For a voice memo or a podcast raw track: whatever the recorder gives you, because the content stops well below any of these ceilings anyway.

The rate is rarely the thing that is wrong. When audio sounds dull, the far more common cause is a lossy stage in its past — which shows up as content stopping early relative to the rate, not as the rate itself.