What is de-essing?

What is sibilance?

Sibilance is the hiss in a voice on s, sh, z, ch and sometimes t. You make those sounds by forcing air through a narrow gap between the tongue and the teeth, and what comes out is mostly noise high up the spectrum, between about 4 and 10 kHz. A sh sits lower than an s, and higher voices tend to have higher sibilance than lower ones.

Some sibilance is part of speech, and you need it to understand the words. It becomes a problem when the esses are much louder than the vowels around them, so every s stings on headphones or jumps out of the mix like a hi-hat. A typical vocal chain makes it worse at almost every stage. Condenser microphones often have a lift somewhere above 5 kHz, and singing close and straight into the capsule adds more. A compressor then evens out the loud vowels and the quieter consonants, which brings the esses up compared with everything else, and the brightening EQ that comes after lifts them again.

Press Trigger in the lab and you'll hear the problem on its own. With everything under 6 kHz filtered away, the vocal loop is mostly quiet, with loud bursts of hiss wherever there's an s. Those bursts are what a de-esser goes after.

How a de-esser works

A de-esser is a compressor with a filter in front of its detector. A normal compressor reacts to the level of the whole sound, so a loud vowel turns it down and a sharp s often doesn't. A de-esser listens to a filtered copy instead, usually through a high-pass or band-pass filter set somewhere between 4 and 10 kHz. That copy is called the sidechain. Vowels have little energy up there and esses have a lot, so the detector barely hears the vowels and hears every s clearly.

When the sidechain goes over the threshold, the de-esser turns the sound down. In the lab, the loudest s in the vocal loop peaks at about −8 dBFS above 6 kHz. With the threshold at −25 dB, that s is 17 dB over. The lab's de-esser takes off 0.9 dB for every decibel over, so on paper that s would lose about 15 dB. The range stops it at 8 dB, which is the most this de-esser will ever take off, and because the gain needs about a millisecond to move, it never quite catches the sharpest peak of the hiss. The deepest cut in the loop comes to 7.6 dB. A quieter s that only goes 5 dB over loses up to 4.5 dB, and anything that never crosses the line is left alone.

De-essers react fast, because an s is short and the gain has to come down before most of it has gone by. The one in the lab reacts within about a millisecond and lets go over about 60 ms, which is quick enough to catch an s and recover before the vowel that follows. Many de-essers don't show attack and release controls at all; they're set for you.

Reading the lab

The big screen shows only the part of the loop above the frequency knob, which is what the de-esser listens to. Pale is that band as it came in; pink is the same band after the de-esser. The dashed line is the threshold, and anything poking above it, shaded pink, is what sets the de-esser off. Under it, the amber lane shows gain reduction, and you can hover over any point to read the exact figure in decibels.

The lower screen shows the same thing by frequency, from 100 Hz on the left to 20 kHz on the right. The dashed amber curve is the detector's filter, and the detector hears whatever sits under it. The pink line is how far each frequency is being turned down; while the loop plays it dips with every s, and when it's stopped it holds at the deepest point in the loop. Beside it, the coach explains your current settings using numbers measured from the loop, and its tiles show the deepest cut, how many bursts the de-esser caught and how much of the loop it spends working.

The meters show the trigger level going into the detector, the gain reduction and the output. Press A/B to switch the de-esser off: the lab sets the original to the same loudness when you do, so the esses are the only thing that changes. You can use your own vocal too. Upload a file or press Record, and nothing leaves your device.

The controls

De-essers have fewer controls than compressors. FabFilter Pro-DS, the Waves DeEsser and Logic Pro's DeEsser 2 all have a frequency and a threshold, and most de-essers add a range and a choice of split-band or wideband.

Frequency

The frequency decides what the detector listens to. Set it where the esses are loudest compared with the rest of the voice. On the vocal in the lab that's around 6 kHz: above that line the esses are tall bursts and the vowels hardly register.

Set it too low and the detector starts hearing the voice itself. At 3 kHz, with the same threshold, the lab's de-esser reacts 12 times instead of 5 and works for about a third of the loop, ducking syllables that had no s in them. Set it too high and the s slips under the filter, so you need a lower threshold to catch it, and then breaths and cymbal spill start setting it off.

Threshold

The threshold is the level the sidechain has to reach before anything happens. Lower it until the loudest esses cross it while the rest of the vocal stays under. In the lab, −25 dB catches five bursts on the vocal loop and leaves the other 92% of it untouched. Take it down to −45 dB and the de-esser works almost the whole time, which is how most over-de-essed vocals start.

A fixed threshold has one weakness. If the singer whispers the verse and belts the chorus, an s in the chorus is far louder than one in the verse, so a threshold set for the chorus misses the verse entirely. Some de-essers deal with this in a relative mode, which compares the high band with the level of the whole voice so that loud and quiet passages get treated the same way.

Range

The range, sometimes called reduction or depth, is the most the de-esser will ever take off, however loud the s is. With the range at 8 dB, the loudest s in the loop comes down by 7.6 dB, just short of the limit, so even a very sharp s can't be pushed into a lisp.

Aim for 3 to 6 dB of reduction on the worst esses. Past about 10 dB, most voices start to sound like they're saying th.

Split-band or wideband

This decides what gets turned down once the detector reacts. A wideband de-esser turns the whole sound down on each s. It's simple and sounds natural when it's only doing a little, but the body of the voice dips with every s. In the lab, wideband mode drops everything under 6 kHz by almost 4 dB on average while it's working, and you can hear the level wobble on words that run straight into an s.

A split-band de-esser turns down only the high frequencies, so the vowel next to the s keeps its level. The lab does this with a high shelf at the same frequency as the detector. The shelf's gain follows the gain reduction, so it's flat while nothing is happening and dips on each s. Split-band is the better default for a lead vocal. Wideband suits a de-esser placed after a lot of brightening, where the whole sound has turned harsh, and it's what you get when you build a de-esser from an ordinary compressor.

Listen

Most de-essers have a switch that plays only the sidechain. Plug-ins call it Listen, Audition or Sidechain listen, and it's the quickest way to set the frequency: switch it on, sweep the frequency until the esses are loud and the rest of the voice has almost gone, then switch it off. The lab's Trigger button does the same.

Some can also play only what they're removing, which is what the lab's Removed button does. You should hear short ticks of s and t in it. If you can follow the words, the de-esser is taking too much.

De-esser vs EQ

An EQ cut is on all the time. Take 8 dB off everything above 6 kHz and the esses come down, but so does the air on every vowel and the breath before each line. The voice ends up duller between the esses, where there was nothing wrong with it.

A de-esser only cuts while an s is happening. On the vocal loop with the threshold at −25 dB, that's 8% of the time, and for the other 92% the top of the voice is exactly as it was. That's why a de-esser can take more off the esses than an EQ could without the vocal sounding dark. Step 8 in the lab switches to a static high shelf at the same frequency with the same 8 dB, so you can hear the difference at matched loudness.

A dynamic EQ sits between the two: an EQ band whose gain moves with the level, the way a de-esser's does. It's a good choice when the sibilance lives in a narrow range and you want to shape the cut. A multiband compressor with only its top band working can do the job as well, though it's more controls than the problem needs.

How to de-ess a vocal, step by step

This order works in any de-esser, and the lesson in the lab follows it.

1. Start with the range at about 6 dB and the threshold at 0Nothing happens yet, which lets you set the frequency without the de-esser getting in the way.
2. Find the esses with the listen switchSwitch to sidechain listen and sweep the frequency between 4 and 10 kHz. Stop where the esses are loud and the vowels have mostly gone.
3. Lower the thresholdSwitch listen off and bring the threshold down until the gain reduction meter moves on the loudest esses and stays at zero on ordinary words.
4. Check what's being removedIf your de-esser has a delta or removed switch, listen to it. It should be short ticks of s and t, and words in there mean it's taking too much.
5. Listen in the mix, then on headphonesSibilance that sounds fine on speakers can still sting in earbuds, so check both at the level people will listen at.
6. Compare with it offThe vocal should sound just as clear, minus the sting. If it sounds dull or lispy, ease the range back or raise the threshold.

Starting settings for common sounds

These are places to begin. Every voice and microphone puts its sibilance somewhere slightly different, so find the frequency with the listen switch each time.

Backing vocals can take more because they sit behind the lead, and when several voices sing the same s at once, the esses pile up. A reverb send can take the most of all, because a tail full of hiss is distracting and the dry vocal still carries the words.

Where the de-esser goes in a vocal chain

The usual place is after the compressor. Compression lifts the quieter parts of a vocal, consonants included, so the esses are at their loudest coming out of it, and a de-esser there catches them at their worst. If the esses are loud enough to make the compressor jump on every s, add a light de-esser before the compressor as well. Two gentle de-essers usually sound better than one working hard.

Put a de-esser before anything that spreads sibilance out. A reverb or delay turns one sharp s into a trail of hiss, so either de-ess the vocal before the send or put a de-esser at the start of the reverb's return. If you add a bright EQ boost after the de-esser, check the esses again, because the boost undoes some of its work. The vocal chain lesson puts all of this in order, and effect order covers the reasoning behind it.

Common de-essing mistakes

Taking too much off

A vocal with every s turned down 12 dB sounds like the singer has a lisp, which is more distracting than a slightly bright s. Keep the reduction on the worst esses under about 6 dB.

A threshold so low it never lets go

If the gain reduction meter never returns to zero, the de-esser has become a wobbly EQ cut on the whole top end. Raise the threshold until the meter sits at zero between the esses.

Setting it with the vocal soloed

On its own, a vocal's esses always sound louder than they will in the mix, where cymbals and guitars share the same range. Find the frequency soloed if you like, but set the threshold with the whole mix playing.

De-essing a vocal that doesn't need it

Pick Vocal 2 in the lab. Above 6 kHz it peaks about 17 dB lower than the first vocal, so at the same settings its loudest s only just reaches the threshold and the de-esser takes off nothing you could measure. When there's no sting to remove, a de-esser can only take away clarity.

Using a de-esser to fix the recording

Very loud sibilance often comes from singing straight into a bright condenser from close up. Angling the microphone slightly away from the singer's mouth, or raising it a little, softens the esses before they're recorded, and the de-esser then only has to tidy up.

Other ways to tame esses

Turning each s down by hand needs no plug-in at all. In any DAW you can cut a short region around an s and drop its clip gain by a few decibels, or draw a dip in the volume automation. It takes longer, but it never reacts to the wrong thing, which makes it the tidiest fix for the odd s a de-esser can't handle cleanly.

You can also build a de-esser from any compressor with a sidechain filter. In Ableton Live, open the sidechain section of Compressor, switch on its EQ and set it to a high-pass or band-pass around 6 kHz. The compressor then reacts to the esses and turns the whole sound down on each one, so what you've built is a wideband de-esser. The sidechain compression lesson explains how sidechains work in more detail.

Training your ears to hear de-essing

Good de-essing is meant to be hard to hear, which makes it hard to learn. The quiz under the lab plays the sound untouched, through a de-esser or with a static EQ cut, all at the same loudness and with the settings hidden, and asks which you're hearing. Listen to the esses first and then to the air between them, because a de-esser leaves the gaps alone and an EQ doesn't. There are eight rounds, and your best score is kept in your browser.

Questions people ask

Keep going