"Soft clipping" is an oxymoron#
Clipping has a definition. I take it from one of my favourite textbooks, Douglas Self's Small Signal Audio Design. Self sets four requirements for a proper clipper. It is symmetrical, with no time component. The top of the waveform is absolutely flat. You can set the threshold. And below the threshold it does nothing at all. In other words, a clamp: if the sample is bigger than the threshold, set it to the threshold, and the same for the negative side. That's it.
So once you round it off before the threshold, it is not a clamp any more, and it is not clipping. It is some other distortion curve, tanh or whatever. The accurate word for a smooth transition towards a hard limit is saturation. Either it is a clamp or it is not a clipper. It is my opinion, regardless of whether anyone agrees with me, that anything sold as a "soft clipper" is simply incorrectly named, and calling analogue parts such as FETs "soft clipping" is equally wrong.
To be fully transparent: Self doesn't reject the term "soft clipping". He only says that soft is not proper clipping, and that a well-designed clipper is flat-topped. The claim that the term is an oxymoron is my own.
Clipping destroys information#
Saturation remaps the sample values. Clipping deletes them. Every value above the threshold becomes the same value, so there is no way back: you don't know what was what.
My Saturation Reverser shows the difference. It undoes tanh saturation and nulls against the original, right up until you push it into actual clipping. Then you hear nasty crackles, because you can't reverse information loss.
Analogue does not clip softly either#
"But in analogue, clipping is soft and round." Well, every analogue device has a hard limit, otherwise you could take a tiny one-watt tube amp and play Wembley Stadium with it without the PA. Like Self says, a well-designed analogue clipper is flat-topped, exactly like digital clipping. What analogue clipping does avoid is aliasing, because there is no sampling frequency. Naive digital clipping aliases heavily; see oversampling.
Music needs its transients#
The popular advice is to clip the kick or snare "just a couple of dB above the peak", because it only costs one or two samples. If it were one or two samples, fine. But real hits all peak at different levels, so you would have to automate every single beat, and when I zoomed in on a clipped kick it was more like a hundred samples.
If you put clippers on every percussive track and one before the limiter "to take the load off", you have just made music without transients. When you loudness-match the result, the mix sounds mushy, lethargic and harsh in the highs.
It does not even make things louder#
People who clip heavily are chasing loud, not good. Normalise a clipped snare against the unclipped one and it sounds weak and feeble next to the original transient. And the listener owns the volume knob anyway (see loudness). And the gain is perhaps 1 or 2 dB. Someone who knows how to mix loud without clippers may well end up louder than someone who clips and doesn't.
The flat tops do not survive the real world#
Any phase rotation undoes them: the all-pass filters in radio chains, car hi-fi, vinyl cutting. In the video, I boosted a master 20 dB into clipping, which made it really loud initially. But as soon as you run it through an all-pass filter, it had to be pulled down about 10 dB to stay below 0 dBFS. On vinyl, cut level is set by groove geometry, not peaks, so clipping for a louder cut just makes the cut worse.
People who say "well my clipped tunes play fine on festival rigs" should know that squared-off waveforms carry sustained power, and sustained power cooks voice coils. The PA manufacturers' own manuals say so. If the PA protection limiters catch it first, they just turn your music down.
A clipper is one line of code#
Paying premium money for a clipper is among the most foolish purchases in audio. I would never use multiband clipping in mastering: done for real, it sounds crackly and horrible, and the expensive "multiband clippers" are actually overlapping multiband saturation you could just build yourself, for example in Dogma. Some good distortion plugins call themselves clippers, and I wish they would call themselves what they are: distortion plugins.
This website uses no cookies directly. However, embedded YouTube videos may set cookies and process personal data once played. By interacting with the video, you consent to YouTube’s data processing. See YouTube’s privacy policy for details.
Tools and courses
References
- 01
Self, D. (2020). Small Signal Audio Design (3rd edition), chapter 24, Level Control and Special Circuits. Focal Press.
Sets out requirements for an ideal clipping circuit, including symmetrical clipping levels that do not depend on frequency or signal history and an absolutely flat top, and notes that the gradual onset of diode clipping makes it soft and so falls short of them.
Supports: "Soft clipping" is an oxymoron - 02
Esqueda, F., Bilbao, S. & Välimäki, V. (2016). Aliasing reduction in clipped signals. IEEE Transactions on Signal Processing, 64(20), 5255–5267.
Hard clipping in the digital domain creates harmonics that are not band-limited, so components above the Nyquist limit fold back into the audible band as aliasing; oversampling is the standard remedy.
Supports: Analogue does not clip softly either - 03
Eastern Acoustic Works (1997). Understanding loudspeaker power handling and selecting the proper power amplifier. EAW application note.
As an amplifier clips, its output approaches a square wave and its RMS level rises towards the peak level, increasing voice-coil heating enough to cause thermal failure at a lower signal level than expected.
Supports: The flat tops do not survive the real world