Clipping & Headroom

Almost every audio problem is fixable after the fact. Too quiet? Turn it up. Hissy? Denoise it. Boomy? EQ it. Clipping is the exception: the waveform's tops are gone, replaced with nothing, and no processor can truthfully put them back. Headroom is the space you keep so that never happens — and how much to keep is one of the oldest settled questions in audio engineering.

What clipping is

Every audio system has a maximum level it can represent. In digital audio that ceiling is 0 dBFS — decibels relative to full scale, the largest number the sample format can hold. Ask the system to record a waveform that swings past the ceiling and it writes the only value it has: the maximum, over and over, until the wave comes back down. On a scope you see a sine wave with its crests sliced flat — clipped, like a hedge.

Those flat tops are new content. A smooth wave abruptly flattening generates a spray of high-frequency harmonics that were never in the source, which the ear reads as crackling, buzzing distortion riding on every loud syllable. And because the original shape of the peak was never stored, the damage is done at capture. Declipping tools can interpolate a guess at what the peak might have been; none of them can know.

Clipping can also strike before the digital ceiling is anywhere in sight. Every stage in the analog front end — the mic capsule at extreme sound pressure, the preamp near maximum gain — has its own overload point, and a signal clipped at the preamp arrives at the converter already flat-topped, reading a deceptively safe −6 dBFS on the meter. That's why gain staging treats headroom as something every stage needs, not just the last one.

Soft shoulders: how analog forgave you

Engineers of the tape era were more relaxed about peaks, and they had reason to be. Analog media overload gradually. Push magnetic tape past its nominal maximum and it saturates: the material's magnetization compresses the peaks smoothly, adding warm-sounding low-order harmonics rather than snapping the waveform off. Tube circuits behave similarly, bending the waveform progressively as they run out of voltage swing. Overload in analog was a shoulder — a region you could lean into, sometimes deliberately, as an effect. Whole genres were built on the sound of tape and tube stages driven past their ratings.

That forgiveness shaped habits. Meters like the VU deliberately averaged over peaks because occasional transients into the red were survivable. When recording went digital — through the pioneering PCM work of the 1970s, including Thomas Stockham's Soundstream recordings in the United States, and then the Compact Disc's 16-bit format in 1982 — engineers carried those habits onto a medium with entirely different physics, and discovered the hard way that digital does not have a shoulder.

Why full scale is a cliff

Digital's ceiling is not a material gradually running out of magnetism; it's an integer running out of bits. One sample below full scale, the system is perfectly linear — distortion essentially unmeasurable. One sample above, it is fully, instantly nonlinear. There is no transition region: 0 dBFS is a cliff, and the character of the distortion past it is uglier than analog's. Hard clipping generates strong high-order harmonics, and in a sampled system the harmonics that land above the Nyquist frequency alias — fold back down as inharmonic tones unrelated to the note that caused them. Tape saturation flatters; digital clipping just breaks.

The cliff even has a subtlety at its edge: a digital signal can measure just under full scale on a sample-peak meter yet exceed it when the DAC reconstructs the continuous waveform between samples. These inter-sample overs are why modern loudness standards such as EBU R128 specify true-peak measurement, and why sensible practice never aims anywhere near the top in the first place.

Headroom: the space you promise not to use

Headroom is the gap between your normal operating level and the ceiling — insurance measured in decibels. Broadcast institutionalized it long ago. European broadcasters set their alignment level at −18 dBFS (EBU R68), leaving 18 dB between reference tone and full scale; American film and television practice aligns at −20 dBFS. Those margins weren't chosen timidly — they reflect measurement of how far real program material spikes above its average when someone laughs, coughs, or slams a script on the desk.

For conversational speech, peaks around −12 dBFS are a well-grounded target. Speech is less spiky than a drum kit, and it's usually been through compression that tames its worst excursions, so 12 dB of margin covers realistic surprises with room to spare. Meanwhile −12 keeps the voice far above the noise floor of any competent 24-bit chain, so the headroom costs nothing. It's the same trade gain staging makes everywhere: don't chase level early; keep margin where damage is irreversible and add level later where it's free.

Protecting a voice in real time

Live speech — calls, streams, recorded podcasts — is where clipping stakes are highest, because there is no second take. Three practices do most of the protective work. Set the input level by measurement rather than eye, so peaks start in a known place (calibration exists precisely for this). Watch a meter that reports peaks, not just average level — RMS and peak can differ by more than 12 dB on voice, and it's the peaks that hit the ceiling. And use a latching clip indicator: a clipped syllable lasts milliseconds, far too brief to catch on a bouncing bar, so the light must stay lit until you've seen it. If the indicator ever fires during a normal sentence, the fix belongs at the gain dial, not in post.

It's tempting to assume the call platform will save you — Zoom, Meet and Teams all run their own automatic gain control. But platform processing sits after your input, and gain reduction applied to an already-clipped signal just gives you quieter distortion: the flat tops scale down along with everything else. Headroom is only useful upstream of the damage, which means it has to be yours.

In DeskBroadcast

DeskBroadcast's meters show RMS and peak in dBFS, with a CLIP indicator that latches the moment peaks exceed −0.5 dBFS and stays lit until you click it to reset — so a clipped consonant during a call can't slip past unnoticed. The processing chain targets −12 dBFS speech peaks, keeping a full 12 dB of headroom in hand at all times. And the protection starts before the meters ever matter: guided calibration warns "Too hot" while you're setting the mic's dial, long before the level gets anywhere near the cliff.