A sample rate captures frequencies up to half of itself — the Nyquist limit — so 44.1 kHz reaches 22.05 kHz, comfortably above the top of human hearing. Doubling to 96 kHz costs twice the storage and delivers headroom that nobody can hear directly. Uncompressed bitrate is sample rate × bit depth × channels, so 48 kHz 24-bit stereo is 2,304 kbps and an hour of it is well over a gigabyte.
That does not make high rates pointless. It makes their benefit indirect, and worth stating precisely rather than assuming.
What does the arithmetic give?
Uncompressed size is sample rate × bit depth × channels × seconds ÷ 8.
| Format | Bitrate | One hour |
|---|---|---|
| 44.1 kHz, 16-bit, stereo | 1,411 kbps | ~605 MB |
| 44.1 kHz, 24-bit, stereo | 2,117 kbps | ~908 MB |
| 48 kHz, 24-bit, stereo | 2,304 kbps | ~988 MB |
| 96 kHz, 24-bit, stereo | 4,608 kbps | ~1.98 GB |
| 320 kbps compressed | 320 kbps | ~137 MB |
Storage scales linearly with all three factors, which makes the decision easy to reason about and easy to get wrong by a factor of four without noticing.
What is the real argument for higher rates?
Filter headroom. A converter needs an anti-aliasing filter below half the sample rate, and that filter cannot be infinitely steep — so at 44.1 kHz it has to do its work in the narrow band between about 20 and 22 kHz.
A higher rate moves that filter far above the audible range, where its phase behaviour and ripple cannot interact with anything you can hear. The same applies to processing: some nonlinear plugins generate harmonics above Nyquist that fold back as aliasing, and running at a higher rate pushes those artefacts out of the way.
Both are arguments about the production chain rather than about the delivered file. Which is why the professional convention is to record and mix high and deliver at 44.1 or 48 kHz.
What does bit depth change?
Dynamic range, not frequency. Each bit is about 6 dB, so 16-bit gives roughly 96 dB of range and 24-bit about 144 dB.
Ninety-six decibels is more than adequate for a finished master. Where 24-bit earns its keep is in recording, where you want to leave substantial headroom below clipping and still have a low noise floor — with 24 bits you can record 20 dB quiet and lose nothing that matters.
It is the same trade as shooting a photograph with room to expose down: the extra depth is insurance during capture rather than quality at delivery.
Channel count multiplies everything and is the easiest factor to forget. A stereo pair is twice a mono file; a 5.1 mix is six times it; and an ambisonic or object-based session can be dozens of channels, at which point the storage arithmetic stops being an afterthought and starts driving the project setup.
Which rate should a project use?
Match the destination, and stay there. Video is 48 kHz almost universally; music distribution is 44.1 kHz; both are converted from whatever the session ran at, and each conversion is a resample.
The one thing worth avoiding is repeated rate changes. Every resample is an interpolation, and while a good one is transparent, a chain of four of them in a workflow is not doing the material any favours.
What does compression actually remove?
Detail the encoder predicts you will not miss, chosen by a psychoacoustic model rather than by frequency. Loud sounds mask quieter ones nearby in frequency and in time, and a lossy codec spends its bits on what remains audible after that masking is accounted for.
That is why the same bitrate sounds very different on different material. Dense mixes hide a great deal; a solo instrument with silence around it hides almost nothing, and it is where low-bitrate artefacts are easiest to hear.
Questions people ask
Is 320 kbps transparent? For most listeners on most material, yes. The audible differences that remain are on specific pathological content — applause, cymbals, some synthesis — rather than across the board.
Does a higher sample rate improve bass? No. Bass is nowhere near the Nyquist limit at any common rate. Low-frequency reproduction is a monitoring and room problem, not a sampling one.
What is Nyquist exactly? The theorem that a signal band-limited below half the sample rate can be reconstructed exactly from its samples. The practical caveat is the word band-limited — anything above the limit has to be filtered out before sampling, not after.
Why 44.1 and not 44? It descends from fitting digital audio onto video tape in the early development of the CD, where the available line and frame rates produced that specific figure. It is an artefact of 1970s hardware that has outlived the hardware by half a century.
Should I upsample an old recording? It gains nothing that was not captured. Upsampling can make a later processing stage behave better, and it cannot recover bandwidth the original never had.
Does dithering matter? When reducing bit depth, yes. Truncating 24 bits to 16 produces correlated quantisation error that sounds worse than the noise dither replaces it with, which is why it is the last step of a master rather than an option.
Rate sets bandwidth, depth sets dynamic range, and neither is the reason a recording sounds good. The sample rate calculator gives the bitrate and the Nyquist limit, and the audio file size calculator turns any of it into megabytes per hour.