There’s a particular kind of harshness — a crackle, a flattened rasp on the loudest words of a sentence — that no amount of editing after the fact can fully undo. Unlike almost every other audio problem, this one is not a matter of taste or polish. It’s a hard physical limit being crossed, and once it’s crossed, the information needed to reverse it is simply gone.
What the ceiling actually is
Every digital audio system has a maximum level it’s capable of representing — a ceiling, conventionally marked as 0 decibels. This isn’t a soft guideline or a quality setting; it’s the literal upper bound of what the numbers in the system can express. A sound that arrives louder than that ceiling doesn’t get recorded as “very loud” — the system has no way to represent a value above its own maximum, so the peak of the waveform is simply cut off flat at the ceiling instead of following its natural curve. That flattening — a smooth wave abruptly truncated — is what the ear interprets as harshness or crackle. It’s not a matter of the sound being unpleasant; it’s a literal distortion of the waveform’s shape.
This is also why clipping can’t be fixed in post-production the way many other issues can. A quiet recording can be made louder. A muddy recording can be equalized. But a clipped waveform is missing information — the actual peak of the original sound was never captured, only a flat-topped approximation of it — and no amount of processing can reconstruct data that was never recorded in the first place. This single fact is what separates clipping from almost every other audio problem: everything else is a matter of degree, fixable with more effort. Clipping is closer to a permanent loss.
Why it happens even when everything “sounded fine”
Clipping is fundamentally a question of headroom — how much louder a sound could get before hitting the ceiling, from wherever its current level sits. Headroom disappears in a few characteristic ways. The most direct is simply setting the input level too high to begin with, so that even ordinary speech sits close to the ceiling and any moment of increased energy — emphasis, a laugh, excitement — pushes past it. A second, less obvious cause is that a person’s distance from a microphone and their vocal energy both vary naturally during a real conversation, and a level that was safe during a calm moment stops being safe the instant someone leans in or raises their voice.
The least obvious cause, and the one that catches even careful people, is that a final mix is often the sum of several sources — a voice, background music, another participant’s microphone — each of which can look perfectly safe on its own meter. The ceiling, however, applies to the combined output, not to each source individually. Several moderate signals summed together can exceed the ceiling even when none of them appeared to be a problem on its own, which is why checking each input separately is not the same as checking the thing that actually gets sent out.
The principle behind fixing it
The fix is fundamentally about preserving headroom rather than eliminating it — setting levels so that normal, expected loudness sits comfortably below the ceiling, leaving room for the moments that are louder than normal, rather than tuning levels for the average moment and hoping the loud ones never arrive. This is a deliberate trade-off: a bit more headroom means a quieter-sounding average level, in exchange for the loud moments never being lost outright.
Because the real risk lives in the unpredictable moments — the laugh, the emphasis, the point in the conversation nobody planned for — a check performed only during calm, predictable speech doesn’t actually verify anything. And because the danger is in the combined output rather than any single input, a check that only looks at individual sources can pass cleanly while the actual broadcast or recording still clips. Catching this reliably means verifying the true, combined signal under realistic, energetic conditions before recording begins — not glancing at a quiet test and assuming the rest of the session will behave the same way.