Two different problems wearing one name
Most plugins describe a single bitcrush control, and that control almost always moves sample rate, bit depth or both together. They are not two settings of one thing. One adds incorrect frequencies, the other adds noise and coarseness. They sit at opposite ends of the frequency domain and they behave differently under level changes.
Any bit depth has a finite number of steps across the full amplitude range. The gap between adjacent steps is the quantisation error, and it behaves as a low-level broadband noise floor that follows the signal. The step size for N bits works out at roughly minus 6.02 times N dBFS, which puts 8 bits near minus 48 dBFS and 16 bits near minus 96 dBFS.
Reducing sample rate brings aliasing back
Drop the rate and the Nyquist ceiling moves down with it. A full-band signal resampled to 8 kHz has a ceiling of 4 kHz, so every harmonic above that folds back into the range you are already using. This is exactly the same mechanism as the one in any aliasing complaint, and a 32 kHz resample on a bright pad produces grit without the producer having done anything unusual.
The staircase you see on the waveform comes from the other operation. It is not related to the folding, and filtering before the reduction only helps the folding. Confusing the two is common enough that it is worth naming: if the artefact moves when you change pitch, it is aliasing, and if it stays constant through a note, it is quantisation.
- Low-pass before reducing so unwanted harmonics are gone rather than relocated.
- Keep rate reduction on parts with little high-frequency content, such as bass, drum bodies and tape-like noise beds.
- Resample to a whole number that divides the session rate, such as 24 kHz from 48 kHz, to avoid a fractional resampler.
- Bounce the affected part at the reduced rate instead of running the whole project down, which keeps the rest of the mix clean.
Reducing bit depth adds noise and a staircase
Depth reduction is a rounding operation. Each sample is snapped to the nearest available step, and the difference between the true value and the snapped one is dither-shaped noise. It does not move frequencies. It makes the waveform blocky, and it puts a floor under the signal that no amount of volume reduction will remove.
| Bit depth | Approximate noise floor | What it sounds like |
|---|---|---|
| 16 bit | -96 dBFS | Inaudible in practice |
| 12 bit | -72 dBFS | Faint grit on quiet material |
| 8 bit | -48 dBFS | Audible hiss and blocky transients |
| 4 bit | -24 dBFS | Lo-fi speech effect |
There is a specific relationship worth knowing. Cutting a stereo file from 16 to 8 bits costs about 48 dB of signal-to-noise ratio, which is far more damage than most people expect from something described as an aesthetic control. If the source is already noisy, the two floors add and the result can become unusable in a busy arrangement.
The cost of using both deliberately
Neither degradation is recoverable, so this is a genuine trade rather than a processing chain to be improved later. Everything you lose is lost permanently, and the honest description of a lo-fi treatment is that it removed information and added information that was not in the original.
What makes it a creative choice rather than damage is placement. A crushed hi-hat in a mix with strong kick and bass needs no clarity to do its job, and the density hides what it removed. The same treatment on a lead vocal removes intelligibility in a way nothing can mask, because there is nothing else competing for that frequency band at that moment.
The split test is simple and worth doing. Bypass the crush and listen to whether the part loses its role or merely loses character. If the arrangement still functions, the treatment is paying for itself. If it needs a corrective EQ to become audible again, the processing cost more than it delivered.
Practical notes for the session
Always keep a clean print. Bitcrushing is one-way, and the temptation to undo it by automating the control is not the same as having the original signal, because the folding and the noise floor both remain in the file the plugin wrote.
On level, drop the signal by roughly 6 dB before crushing. Coarse steps against a near-ceiling signal clip, and that clip is a separate distortion stage nobody chose. Checking the result on a meter costs nothing and settles the question immediately.
The bottom line
Sample rate reduction and bit depth reduction damage the signal in completely different ways, and tools that offer one knob labelled bitcrushing are usually doing both. Rate reduction reintroduces aliasing, which is wrong information added; depth reduction adds quantisation noise and a staircase, which is coarse information. Both are legitimate creative choices with a real cost, and the cost is only worth paying where the arrangement hides it.
Judge the artefacts in context
432Hz MASTER gives you a fixed reference pitch, so you can hear whether degradation added character or removed information.
Frequently asked questions
Does lowering the sample rate just remove high frequencies?
No, it moves them. Content above half the new rate cannot be represented and folds back into the audible band as wrong harmonics. Use a low-pass before the reduction if you want them gone rather than relocated.
What does bit depth reduction actually add?
Quantisation noise, spread across the frequency band at roughly the level of the least significant bit. At 8 bits that floor sits near minus 48 dBFS and is plainly audible; at 16 bits it is near minus 96 dBFS and generally inaudible.
Why does my crushed sound get quieter?
Lower resolution reduces peak headroom and the coarse steps clip against the ceiling, which usually triggers a limiter or an output stage. Drop the level before crushing or compensate after, and check with a meter rather than by eye.
Can bitcrushing be reversed?
No. Once folded or quantised, the information is gone and no algorithm can reconstruct which frequencies were there. Keep a clean print before processing so you can always return to it.