The complete guide to de-essing

The de-esser might be the most misunderstood tool in the vocal chain. Everyone knows what it's for: taming the harsh ess sounds that make a vocal spit and sizzle. But almost nobody is taught how it actually works. So most of us learn it by ear, badly, and end up dulling the very thing we were trying to protect. This is the complete guide we wish we'd had: what sibilance really is, how de-essers work, how to de-ess without killing the air, and when the tool you actually need isn't a de-esser at all.

We build one, so we've spent an unreasonable amount of time on this. Where we make a claim about our own plugin, it's a measured number you can check for yourself. Everything else is craft.

Key Takeaways

• Sibilance is a set of phonemes (s, sh, z, ch, t, and harder f/th), not a fixed frequency. It usually lives between roughly 5 kHz and 10 kHz, but the exact spot moves with the singer, the mic, and the word.
• Good de-essing does two things: it acts only on the ess, and only while the ess is happening. Break either rule and you get a dull vocal or an obvious pump.
• A static EQ cut can't de-ess: sibilance is intermittent, so a permanent cut is dull all the time to fix a problem that's there some of the time.
• Set the amount on the worst word, then check that held vowels stay untouched. The single best move is to solo what the de-esser is removing.
• If harshness is spread across the whole spectrum, you want dynamic EQ or a resonance tool, not a de-esser.

What is sibilance, really?

Sibilance is a group of consonant sounds, not a band on the spectrum: the consonants you make by pushing a jet of air through your teeth and tongue — s, sh, z, ch, t, and the harder f and th. On most voices their energy concentrates somewhere between 5 kHz and 10 kHz, but that range is a starting point, not a setting. A bright singer might peak up at 8 or 9 kHz; a deeper voice closer to 5 or 6. It even shifts word to word.

That single fact explains most de-essing failures. If you treat sibilance as a fixed frequency and park a dip there, you'll miss every ess that lands somewhere else and gouge the ones that don't. The target keeps moving, so the processing has to move with it. (We go deeper on this in how to de-ess vocals without dulling them.)

Why do vocals get harsh in the first place?

Harsh esses aren't a moral failing of the singer. They come from a stack of ordinary causes: a bright condenser microphone, close mic placement, a naturally sibilant voice, de-essing-unfriendly diction, and then everything you do after — the high-shelf boost for "air," the compressor that lifts quiet consonants, the saturation that adds high harmonics, the limiter that squashes everything up against the ceiling. Each step is reasonable on its own. Stacked, they turn a gentle s into an ice pick.

Understanding that chain matters because it tells you where to fix the problem. Sometimes the honest fix is earlier — a different mic angle, a de-esser before the bright compressor instead of after it — not a heavier hand at the end.

How does a de-esser actually work?

A de-esser is a compressor with a filtered ear: it listens to a high band, and when the energy there crosses a threshold, it pulls the gain down. Two design choices decide whether that's transparent or destructive.

Wideband vs split-band. A wideband de-esser ducks the entire signal when it fires, so a loud ess dims the whole vocal for a moment. A split-band de-esser only reduces the high region where the ess lives, leaving the body of the voice alone. Split-band is almost always the better starting point, and the narrower the split, the more air survives.

What triggers it. The cheap approach fires on level: anything loud in the high band trips it, including a brightly sung vowel, a breath, or a cymbal bleeding into the vocal mic. The better approach tries to recognise that the loud thing is actually a consonant before it acts. That difference — a level trigger versus one that classifies the sound — is the whole ballgame between a de-esser that's transparent and one that just ducks your top end.

This is the part we obsessed over. Sibilance uses a phoneme-aware detector that classifies and protects consonants across several bands, rather than reacting to raw level. In our own gated test harness that comes out to roughly 5 dB of sibilance removed while the non-sibilant high frequencies stay within ±0.04 dB of untouched. We publish those numbers, and the 146 checks and 22 gates behind them, on the validation page — because "transparent" should be a measurement, not an adjective.

The two rules of good de-essing

Strip away the marketing and every transparent de-esser obeys two rules:

1. Act only on the ess. Not the vowel, not the breath, not the cymbal. If your de-esser dims a held, open vowel when the singer leans in, it's chasing brightness, not sibilance, and it will make the vocal dull.
2. Act only while the ess is happening. Sibilance is a flicker — a few milliseconds per word, then gone. The processing should clamp down for exactly that flicker and get out of the way the instant it's over.

Get both right and the vocal comes out smooth and open. Break the first and you dull the voice; break the second and you hear the de-esser breathing. There is no third secret.

How to de-ess vocals, step by step

Here's the workflow we'd teach a friend, in order:

1. Find the ess. Solo the vocal and listen for the words that bite. If you can watch a spectrum analyser, the esses show up as little bursts up top. Note roughly where they sit — it tells you where to point the de-esser.
2. Set the amount on the worst word. Pull the reduction down until the harshness leaves the hardest ess in the take, then stop. Tuning to the worst case keeps you from over-processing the ninety percent of esses that were already fine.
3. Check the vowels. Play a phrase with long, open, sung vowels. They should be completely untouched. If they dip when the singer opens up, back off or narrow the band — you're triggering on brightness.
4. Listen to what you're removing. This is the move that changes everything: solo only the signal the de-esser is taking out. You want to hear esses and almost nothing else. If you hear whole words, tone, or breath, you're removing too much. (Sibilance calls this Δ Listen; most plugins have a "listen" or "diff" mode. Use it every single time.)

If you want the long version of the "don't dull it" philosophy, we wrote a whole piece on de-essing without dulling the vocal.

Where to put the de-esser (and a note on latency)

As a rule, de-ess before anything that adds high-frequency energy — bright compression, saturation, an air shelf. De-essing early means those later stages aren't amplifying an ess you could have tamed for free. Some engineers run a light de-ess before and after; that's fine, as long as each pass is gentle.

One technical wrinkle worth understanding is look-ahead latency. Letting the de-esser see an ess coming catches the onset cleanly, but every millisecond of look-ahead is latency you pay across the whole session. More isn't better — a few milliseconds already beats the transient. We explain the tradeoff, and why we run about 5 ms in the studio and near-zero live, in what look-ahead latency actually buys you.

Common de-essing mistakes

• Reaching for a static EQ cut. Sibilance is intermittent; a permanent notch is dull all the time to fix a problem that's present some of the time. De-essing is dynamic on purpose.
• Over-de-essing. Past a point the esses stop sounding harsh and start sounding lisped. If your singer suddenly has a speech impediment, you've gone too far.
• Chasing brightness instead of sibilance. If held vowels dim, your detector is triggering on level, not the consonant. Narrow the band or switch to a smarter tool.
• De-essing the whole mix bus. A de-esser is a vocal (or per-track) tool. On a full mix it'll duck cymbals, hi-hats, and air along with the esses.
• Setting and forgetting a frequency. The ess moves. If your de-esser lets you lock one frequency, you'll be wrong on half the words.

When you don't actually need a de-esser

Sometimes the harshness isn't sibilance at all — it's a broadband brittleness, a resonant peak, or a room ring smeared across the whole top end. A de-esser is the wrong shape for that job. Reach instead for a dynamic EQ (to tame a specific resonant frequency only when it flares) or a broadband resonance suppressor (to hunt harshness across the spectrum).

If what you're fighting really is sibilance, though, a focused, measured de-esser is the tool built for the exact shape of the problem. We put Sibilance head-to-head against the reference options — including FabFilter Pro-DS — so you can see where a dedicated de-esser leads and where it doesn't.

Frequently asked questions

What frequency is sibilance?

Usually between about 5 kHz and 10 kHz, but there's no single correct number. Brighter voices sit higher, deeper voices lower, and it moves word to word — which is why a moving, phoneme-aware detector beats a fixed-frequency cut.

Is a de-esser the same as an EQ?

No. An EQ change is static — it's on all the time. A de-esser is dynamic: it only reduces the sibilant region while an ess is actually happening, then releases. That's the whole reason it can tame harsh esses without dulling the rest of the vocal.

Can I de-ess with a free plugin?

Yes. A good free de-esser is plenty for most vocals. Sibilance Open is a full, commercially-licensed free tier, not a demo with the best parts switched off — a reasonable place to start before deciding you need anything more.

Should I de-ess before or after compression?

Generally before any stage that adds high-frequency energy (bright compression, saturation, an air shelf), so those stages aren't amplifying an ess you could have handled first. A gentle second pass after is fine.

The short version

De-essing isn't complicated once you see it clearly: sibilance is a moving set of consonant sounds, not a frequency; and taming it transparently means acting only on the ess, and only while it's happening. Set the amount on the worst word, protect the vowels, and always listen to what you're removing. Do that and your vocals come out smooth without going dull.

When you want to try it on your own voice, Sibilance has a genuinely free tier and a 7-day Pro trial with no card — the surest test is your own ears on your own take.

Sibilance is the single-authority de-esser this comes from — a real free tier, and a 7-day Pro trial with no card.

Related reading