De-essing vocals: how to find sibilance and tame it without a lisp

Sibilance is the hiss in "s", "sh", "z" and sometimes "t" and "ch". Every voice makes it, and a vocal needs some of it to be understood. The trouble starts when those sounds come out louder and sharper than the vowels around them, so every "yes" and "special" lands like a small needle. It gets worse as you add the things that make a vocal sit forward: a bright condenser mic, a high-shelf boost, heavy compression that lifts quiet consonants.

A de-esser is the usual fix. It is a compressor that only listens to a narrow, high band of the signal and only acts when that band gets loud. This guide covers how to find the problem, how to set one up, and when to reach for something else. Where the de-esser sits in a chain is covered in our vocal chain basics; this piece is about the job itself.

Step 1: Confirm it really is sibilance

Before you load anything, listen at a moderate level on more than one system. Headphones and small bright speakers exaggerate top end, so a problem you hear only on one pair may not be a problem. Our guide to mixing on headphones explains why it pays to check.

Then rule out lookalikes:

• Harshness across the whole vocal (a constant edge around the upper mids) is a tone problem. Fix it with EQ first; our EQ basics guide covers finding a resonance with a narrow sweep.
• Clicks, lip smacks and breaths are editing problems, not sibilance.
• Distortion on "s" sounds that is baked into the recording will not be fixed by turning them down. You can make it quieter, not clean.

If the vowels sound fine and only the hissy consonants stab out, you have sibilance.

Step 2: Find the frequency

Sibilance does not sit at one fixed frequency. It moves with the singer, the mic and the word. FabFilter's Pro-DS manual gives a useful starting point: normal vocal "s" sounds usually sit around 8 to 10 kHz, and its detection filters can be set anywhere from 2 kHz to 20 kHz. Lower, darker voices and "sh" sounds often sit lower than that.

The fastest way to find it:

1. Loop a phrase with obvious "s" sounds.
2. Turn on the de-esser's sidechain listen (Pro-DS calls it Audition Sidechain; most de-essers have something similar). You now hear only the band the detector is listening to.
3. Sweep the detection band until the "s" sounds are loud and clear in that solo signal and the vowels nearly vanish.

If your de-esser has no listen button, put a narrow, heavily boosted EQ bell on the vocal and sweep it until the "s" sounds jump out. Note that frequency, delete the boost, and use it for detection.

Step 3: Set threshold, then range

The threshold decides when the de-esser acts. Lower it until gain reduction appears on the "s" sounds and nowhere else. Many de-essers offer a second listen mode for exactly this: Pro-DS's Audition Triggering lets you hear which parts of the audio are being turned down, so you can check it catches every "s" and leaves "f", "th" and bright vowels alone.

Range, or depth, caps how much it can turn down. Start small, around 3 to 6 dB of reduction on the worst syllables, and raise it only if the hiss still pokes out. That figure is our working habit, not a rule from any manual; trust your ears over the meter.

Then do the test that matters. Bypass the de-esser and play the phrase again. If the "s" sounds now seem dull or the singer seems to lisp ("yeth" for "yes"), you have gone too far. The aim is consonants that are still crisp, just no longer louder than everything else.

Step 4: Choose wide-band or split-band

This is the one setting that changes the character most. FabFilter's advanced controls page explains the difference plainly:

• Wide-band (full-band) turns down the whole signal for the length of the "s". It is simple and, on a single vocal, often sounds the most natural, because the tone of the consonant does not change, only its level.
• Split-band turns down only the high frequencies. It suits more complex material, such as a full mix or a vocal with music bled into it, because it leaves the lows alone. FabFilter notes that it adds some latency.

If wide-band audibly ducks the body of the voice on every "s", switch to split-band. If split-band makes the "s" sound dull or filtered, go back.

Step 5: No dedicated de-esser? Build one

Any multiband compressor or dynamic EQ can do the job. Ableton's Live 12 manual gives a recipe for its Multiband Dynamics device: enable only the upper band, set its crossover to around 5 kHz, then bring in threshold and ratio gradually for gentle downward compression, with fairly fast attack and release. Soloing the band, as the manual suggests, is the same trick as the sidechain listen above.

A dynamic EQ bell at the frequency you found in step 2, set to cut only when the band crosses a threshold, works the same way and lets you target a narrower slice.

Step 6: Fix the worst ones by hand

A de-esser is an average. It treats a gentle "s" and a vicious one with the same settings, so either the loud ones still stab or the quiet ones get crushed. The cleanest fix for a handful of offenders is manual: cut the "s" into its own region and lower its clip gain or volume automation by a few dB. It takes ten minutes on a three-minute vocal and sounds like nothing at all, which is the point. Our guides to vocal editing and comping and automation in your DAW cover the mechanics.

Doing this before the compressor also helps: a compressor reacts less to an "s" that is no longer a peak.

Common mistakes

• De-essing in solo only. Sibilance that seems huge in solo can be fine in the mix, where cymbals and synths share that range. Make the final call with the track playing.
• One de-esser doing everything. If you need heavy reduction, two gentle stages (or manual edits plus a light de-esser) usually sound better than one hard one.
• Ignoring the reverb. A bright reverb or delay fed by a sibilant vocal smears every "s" into a hiss. De-ess before the send, or put a de-esser on the effect return.
• Boosting air after de-essing without rechecking. A late high-shelf can undo the work.

What we think

De-essing is one of the few mix moves where the best result is the one nobody notices, and that makes it easy to overdo. The default reflex, a preset on every vocal pulled down until the meter moves, quietly sands the edges off a singer. Consonants carry character: the slight whistle in one voice, the soft lisp in another, the hard attack of a rapper's "t". Those are part of how a listener recognises a person rather than a sound.

So treat sibilance as a level problem on specific syllables, not a tone problem with the whole voice. Find the few that hurt, fix those first by hand, then let a gentle de-esser catch the rest. Better still, fix it at the source next time: a mic pointed slightly off the mouth, or a darker mic for a bright singer, often removes most of the need. A vocal that still sounds like the person who sang it is worth a few minutes of clip-gain editing.

FAQ

Where should the de-esser go in the chain? A common order puts it after EQ and compression, so it works on the vocal as it will finally sound. Our vocal chain basics guide explains why, and when to move it earlier.

Can I de-ess a whole mix? Yes, with care. FabFilter recommends split-band processing and its Allround detection mode for full mixes, so the lower frequencies stay untouched.

Is EQ alone enough? Rarely. A static cut dulls the whole vocal all the time; sibilance only happens for a few milliseconds per word, which is why a dynamic tool fits it better.