Audio clipping: what causes it and how to prevent it

Every digital audio system has a hard ceiling. Once a waveform reaches it there is no number available to describe anything louder, so the peaks that would have gone higher are simply cut off flat. That flattening is clipping, and the name comes from the shape it leaves behind: rounded peaks with their tops sliced away.

What you hear is not simply loudness. Squaring off the top of a wave introduces harmonic content that was never in the original sound, which the ear reads as a crackle, a fuzz or a harsh edge on the loudest syllables. On speech it usually shows up first on plosive consonants and on laughter.

Clipping is one specific kind of distortion rather than a synonym for it. Distortion is the broad family; clipping is the member of it caused by hitting a ceiling.

Hard clipping is the abrupt version. The waveform hits the ceiling and stops dead, leaving a flat plateau with sharp corners. This is what a digital converter does when you feed it too much signal, and it is the harsher of the two because those sharp corners generate a lot of unpleasant high-frequency content.

Soft clipping rounds the peaks off gradually instead of cutting them square. Analogue equipment tends to behave this way as it is pushed, and some plugins imitate it deliberately because the result can sound warm rather than broken. It is still distortion; it is simply distortion that many people find pleasant in small amounts.

For spoken word neither is a goal. The difference matters mainly because it explains why one overloaded recording sounds merely thick while another sounds destroyed.

At the microphone preamp. Gain set too high for how loudly someone speaks is the most common cause, and it is the worst one, because the damage is baked into the file before anything else touches it.

At the interface or device input. Some interfaces have their own input stage that can overload even when the software meter looks acceptable. If the hardware has a clip light, believe it over the screen.

Inside the edit. Stacking gain, compression with generous make-up gain, and an equaliser boost can push a track past the ceiling even though the original recording was clean. Check the master meter after you add effects, not only before.

At export. Encoding to a lossy format can produce peaks slightly higher than the ones you saw in the editor, so audio that sat exactly at the ceiling in your project can clip in the file you upload.

Look at the waveform zoomed in. Healthy speech peaks come to a point. Clipped speech peaks arrive at a flat line and sit there for a stretch of samples. Once you have seen the difference at high zoom it is unmistakable.

Watch the meters during the take rather than after it. A clip indicator that lights once during a warm-up is worth stopping for; the same indicator discovered during the edit costs you a re-record.

Listen on headphones to the loudest moment you have. Laughter, a raised voice and the start of a hard consonant are where clipping first becomes audible, and they are easy to skip past when you are scanning a long recording visually.

Aim to have your loudest peaks land somewhere around -12 to -6 dBFS while recording. That leaves several decibels of space above the loudest thing you expect, which is exactly the margin a surprise laugh needs. Recording close to the ceiling buys you nothing: digital audio has plenty of resolution at moderate levels, and you can raise the whole thing later without penalty.

Set the final level at the end of the process rather than the start. Record conservatively, edit, and then bring the finished episode up to a consistent target with a normalizer as the last step.

Put a limiter before the ceiling if your material is unpredictable. A limiter set a decibel or so below full scale will catch the one peak you did not plan for without touching anything else.

Clipping removes information. The part of the waveform that went above the ceiling was never written to the file, so no tool can recover what was there; repair tools work by estimating a plausible curve across the flat section and rebuilding it.

That estimation works reasonably well when the clipped stretches are short and occasional. A handful of clipped consonants across an interview can usually be made unobtrusive. It works poorly when a recording was made far too hot throughout, because there is no clean neighbouring material for the tool to reason from.

If the clipping is mild and the conversation is worth saving, an audio enhancer is worth trying before you write the take off. If it is severe and the speaker is available, re-recording will always beat repair.