The short answer
A microphone hears sound as a smooth, continuous wave. A computer cannot store something continuous, so it takes rapid snapshots of that wave and writes down the value of each one. Each snapshot is a sample, and the sample rate is simply how many of those snapshots it captures every second.
Sample rate is measured in hertz (Hz), which just means times per second, and audio rates are large enough that we usually write them in kilohertz (kHz), or thousands of times per second. So 44.1 kHz means 44,100 snapshots a second and 48 kHz means 48,000 a second. Those two values are by far the most common, and the whole 44.1 versus 48 question is really just about which of them to use and why.
How sampling works
There is a picture most people have here, and it is worth taking apart because it leads somewhere wrong. You imagine the samples as dots, with a line drawn between them, so more dots looks like a smoother line and a truer copy. That is not what happens. Nothing joins the dots with straight segments, and the smoothness of your trace is not what is being measured.
What actually happens is stricter and, once you see it, simpler. Provided the sound contains nothing above half the sample rate, there is exactly one wave that passes through those points. Not a close fit chosen from many, but one. That is the Nyquist and Shannon sampling theorem, and it is a proven result rather than a rule of thumb. Reconstruction is not an approximation that gets better as you add samples, so adding samples does not get you closer to the original. You were never far away.
So a sample rate decides one thing: the highest frequency it can carry, which is half the rate. That is the ceiling. 44.1 kHz reaches about 22 kHz, 48 kHz reaches 24 kHz, and human hearing runs to roughly 20 kHz, with age and loud rooms taking the top of that first. Both common rates already put the ceiling above everything you can hear.
That is also why the two rates sit where they do. A rate a little above 40 kHz covers the whole audible range with room to spare for the filter that removes anything above the ceiling before it becomes numbers, which is the one part of this that genuinely used to be hard. On early converters that filter had to be steep and close to the edge of hearing, and a bad one was audible, so recording higher moved it somewhere harmless. Modern converters are good enough that it mostly does not matter. When a higher rate does sound different, that is the design of that particular converter rather than anything inherent in sampling faster.
What happens above the ceiling
The ceiling is not a soft limit where sound above it simply fades out. A frequency higher than half the sample rate cannot be represented at that rate at all, and instead of going missing it comes back somewhere else. It folds down and appears as a lower frequency that was never played. That is aliasing, and it is the reason the ceiling has to be enforced rather than merely observed: an unwanted 25 kHz tone recorded at 44.1 kHz does not stay inaudible, it arrives inside your voice as a whistle at a frequency nobody produced.
So converters put a filter before the conversion, an anti-aliasing filter, whose only job is to remove everything above the ceiling before the sound becomes numbers. This is the one part of the story where recording higher genuinely used to help, and it was never about your ears. That filter has to be steep, and on early converters building a steep one right at the edge of hearing was hard; a bad one was audible in the range you do listen to. Recording at a higher rate moves the filter far above anything you can hear, so even a mediocre one stops mattering. Modern converters are good enough that this mostly is not a problem any more, and when a higher rate does sound different, that is the design of that particular converter rather than anything inherent in sampling faster.
There is one place aliasing still bites, and it is not at the microphone. Anything non-linear creates frequencies that were not in the signal it was given: saturation, distortion, hard clipping, a guitar amp. Those new harmonics are made after the input filter, so no filter at the front can catch them, and any of them above the ceiling fold back down into your audio as tones that do not belong to the note that caused them. It is the harshness people describe as digital and blame on the sample rate.
The fix lives inside the effect rather than in your session settings. The processing stage runs at a multiple of the rate, where there is room above for the new harmonics, gets filtered, and comes back down. In Patchd's saturator that is a 4x oversampling switch that wraps the waveshaping stage with an anti-alias filter on the way back. It is off by default, because it costs a little latency, and it is the control to reach for when a heavy drive setting starts to sound brittle. Raising the session rate to chase the same problem is the expensive way round.
44.1 kHz versus 48 kHz
The two rates come from two different worlds. 44.1 kHz became the standard for the audio CD, so it is baked into a huge amount of music and consumer audio. 48 kHz became the standard for video, film, and professional audio gear, so it is what most cameras, editors, and interfaces default to today. Neither is higher quality than the other in any way you would notice; they are two conventions that both comfortably cover what you can hear.
Because the audible difference is nil for normal listening and voice, the practical reason to care is matching, not fidelity. When audio at one rate has to play through a device set to another, your system has to resample it, converting 44.1 kHz into 48 kHz or the reverse on the fly. Good resampling is inaudible, but a mismatch that is set up wrong is a classic cause of pitch and speed errors or a steady crackle. The cure is almost always to set everything in the chain to the same rate. You can see what your own system reports with the sample rate checker.
- 44.1 kHz. The CD rate. A safe, universal choice for music you are producing to share as audio files, and still perfectly capable of the full range of human hearing.
- 48 kHz. The video and pro-audio rate, and the default on most modern interfaces and operating systems. The natural pick for anything involving video, streaming, or calls.
Which sample rate should I use?
The honest answer is to match your rate to the company it keeps rather than chase a bigger number. Pick the rate everything else in your project is already using, so nothing has to be resampled.
- Anything with video, streaming, or calls. Use 48 kHz. It is the default across video and most operating systems, so matching it keeps the whole path aligned.
- Music bound for audio files. 44.1 kHz is the traditional home, though 48 kHz is completely fine too. The key is that every track and device in the project agrees.
- Higher rates like 96 kHz. These double the samples, the file sizes, and the processing load in exchange for detail above what you can hear. They have their uses in heavy production, but for voice and listening they mostly just cost more for no audible gain.
Sample rate is one of two numbers that describe a digital audio stream, and they answer different questions. Sample rate sets how high a frequency can be carried at all, which is the ceiling above. The other number, bit depth, sets how finely each measurement is recorded, which shows up as the noise floor underneath the sound rather than as detail within it. They are easy to confuse, so it is worth reading them side by side.
Sample rate in Patchd
In Patchd, a Windows virtual audio mixer, you do not pick a sample rate inside the app. Its engine processes internally at 48 kHz, the common studio default, resamples other device rates to that engine rate, and shows that 48 kHz live in its master clock readout, so you can see the rate the whole mix runs at rather than guess.
Because higher rates cost more CPU for detail you cannot hear on voice, there is rarely a reason to push past your device default here. The engine is real time and runs over ASIO and WASAPI Exclusive, the low-overhead Windows audio paths, adding only single-digit milliseconds of its own delay at a small ASIO buffer. If you care about that delay, the number you actually feel comes mostly from your buffer size rather than your sample rate; see what is monitoring latency for the full picture.
The takeaway
Sample rate is how many snapshots per second capture a sound, written in kHz, and what it sets is the ceiling: the highest frequency that can be captured at all. 44.1 kHz and 48 kHz are the two everyday choices, and both already cover the full range of human hearing, so the difference that matters is not quality but keeping every device and file on the same rate. Match your rate to your project, leave the exotic high rates for the rare jobs that need them, and let your device clock lead.
Patchd is in open beta, with a real-time engine that follows your device's sample rate and shows it live so your audio stays aligned and clean. If you want a Windows mixer that keeps the plumbing honest, join the waitlist and get notified when it is ready. The paid Studio tier is $5.99 a month, $39.99 a year, or $119.99 once for a lifetime license, and the engine is identical on both tiers.