It is sampling a sound much faster than you really need, then shrinking it back down.
Sample x8 fast, let distortion's new harmonics land above the high Nyquist, filter them off, then shrink back to 48 kHz alias-free.
What it is
Running a converter or plugin at a multiple of the normal sample rate, doing the maths, then shrinking back down.
Key facts
Sample rate = samples captured per second; 44.1 kHz = 44,100 samples/sec (CD), 48 kHz = standard for video and live.
Nyquist frequency = sample rate divided by 2; at 48 kHz the highest capturable frequency is 24 kHz.
Oversampling factor x2/x4/x8/x16 means run internally at that multiple: 48 kHz x8 = 384 kHz inside the plugin.
Human hearing tops out near 20 kHz, so you only NEED ~40 kHz; the extra rate is headroom for clean maths.
Aliasing = fake tones created when frequencies above Nyquist fold back DOWN into the audible band as wrong pitches.
Oversampling pushes the Nyquist limit way up (48 kHz x4 = Nyquist 96 kHz) so harmonics land far above hearing, not folded back.
Non-linear processes (distortion, saturation, clipping, limiting) GENERATE new harmonics above Nyquist; that is exactly what aliases.
After processing, a steep anti-aliasing low-pass filter removes the high junk, then decimation drops back to base rate.
Cost: roughly proportional to the factor; x8 = ~8x the CPU of x1 for that plugin.
Linear processes (clean EQ, gain, delay) gain little from oversampling; save it for distortion and limiters.
How it works
Plugin upsamples incoming audio by inserting samples (e.g. x4) and interpolating.
It runs the heavy non-linear maths at the high internal rate.
New harmonics now land well above the original Nyquist, not folded back.
A steep low-pass filter strips everything above the original Nyquist.
Audio is decimated (downsampled) back to the session rate, alias-free.
Real examples
A limiter set to -1 dBTP with oversampling ON catches true-peak overshoots a normal-rate limiter misses.
A guitar amp/distortion plugin sounds smooth instead of fizzy/gritty once x4 oversampling is enabled.
A clipper on the mix bus stops adding harsh digital edge when oversampled.
Bouncing a master: switch the limiter to its highest oversampling (HQ) mode only for the final render.
A saturation plugin on vocals loses the brittle 'spitty' top end with oversampling engaged.
How it helps in live sound
Live = low latency rules: keep oversampling OFF or low (x1/x2) on FOH limiters to avoid added delay.
Reserve high oversampling (x8/x16) for the studio bounce, not the live show.
On a digital console, your DSP runs at a fixed rate; oversampling lives in plugins, not the desk path.
Watch CPU/DSP load: each oversampled plugin can cost several times a normal one mid-show.
For broadcast/streamed events, use a true-peak limiter (-1 to -2 dBTP) which relies on oversampling to read inter-sample peaks.
Everyday analogy
Like filming at 240 frames per second so a fast move stays crisp, then exporting back to normal 24 fps for delivery.
Watch out
Myth: 'oversampling makes everything sound better.' Truth: it only helps non-linear processes (distortion/limiters); on clean EQ or gain it just burns CPU and adds latency.
Fun fact
A signal can clip a DAC even when every sample reads below 0 dBFS, because the real waveform peaks BETWEEN samples; oversampling is how true-peak meters catch these inter-sample overshoots.
Key takeaways
Oversampling = work at a higher internal rate, then shrink back down.
Its main job is killing aliasing from distortion, saturation and limiters.
Higher Nyquist = new harmonics stay above hearing instead of folding back as wrong tones.
Costs CPU and latency, so use the high modes on the final bounce, not live.
Pointless on clean linear EQ/gain; gold on anything non-linear.