De-Esser Settings for Rap (And Where It Goes in the Chain)

Start with a split-band de-esser between 5.5 and 8 kHz, pulling 3 to 6 dB on the loudest esses and nothing on the rest of the take, placed after subtractive EQ and before your first compressor. Deep voices often want 4.5 to 6 kHz, brighter voices 6.5 to 8 kHz. If you saturate the vocal later, add a second light pass in front of the saturator.

Disclosure: we build audio plugins at Ghostnote, and one is a de-esser. It gets one labeled section near the end, and where we quote our own measurements we say so. Everything else here works in the de-esser your DAW already ships with. No affiliate links.

How do you find your sibilance frequency in 30 seconds?

Loop the two worst bars, usually the hook, where the delivery pushes hardest.

1. Insert an EQ first in the chain. One bell, +10 dB, Q around 5.
2. Sweep slowly from 3 kHz to 12 kHz while the loop plays.
3. Stop where the esses turn into an ice pick. Note the frequency; many voices have two.
4. Remove the boost. Set the crossover there or a few hundred Hz below, so the band holds the whole fricative, not just its top half.

Two patterns dominate: a 3 to 5 kHz harshness on sh and ch, and a 6 to 10 kHz spit on s, ts and z. One de-esser near 5.5 kHz splits the difference; two instances, one per band, are cleaner when both are bad.

What de-esser settings should you start with for rap?

Four controls do the work, whatever your plugin calls them. Stock de-essers usually expose the first three; dedicated ones cover the same ground under other names. Where release is hidden, the plugin picks it for you.

• Crossover. Where the detector listens and, on a split-band design, where the reduction lands. Set it from the sweep.
• Threshold or sensitivity. How loud a fricative has to be before the plugin moves. Absolute designs are set in dBFS: park it 4 to 10 dB under the loudest ess. Relative designs compare high-frequency energy to the full-band level, so one setting covers a quiet verse and a loud hook. If yours is absolute, aim for the peak-reduction column below.
• Range. A hard ceiling on how much it may take. Start at 4 dB.
• Attack and release. Attack, where it is exposed, wants a millisecond or so, because the gain has to be down before the fricative starts. Release sets how fast the band comes back.

A split-band de-esser reduces the high band only and sums it with an untouched low band, so the body of the voice is out of reach by construction. Fourth-order Linkwitz-Riley sections sum flat through the crossover, so the split leaves no dent when idle (Rane's primer on Linkwitz-Riley crossovers explains the geometry).

Release is the control most people leave wrong. Too long on a rapid-fire flow and the high band stays clamped through the following vowel, so the bar sounds covered. Too short on a held sung ess and you hear the band bounce. Program-dependent release handles both.

Stacked layers need their own math. Six copies of one ess read like a closed hi-hat, so take 2 to 3 dB per layer plus 2 to 3 dB on the group bus, not 10 dB on the bus alone.

Where does a de-esser go in the vocal chain?

After subtractive EQ, before your first compressor. That placement matters more than any number above.

A compressor's detector reads level, not frequency. A loud ess at 7 kHz looks like a transient to a fast peak detector, so the compressor takes 4 to 6 dB off the whole word to fix a problem that only ever lived in one octave. You hear the syllable dive, then breathe back in. De-ess first and the compressor responds to the voice. The same logic reaches the master limiter, where an untreated ess eats headroom the 808 needed.

A second de-esser usually sits in front of saturation. Drive generates harmonics of whatever you feed it, and fricatives already occupy the region those harmonics land in, so a controlled ess can come out sharp again. One to two dB is plenty there, since the first pass did the work.

Nobody listens to a reverb tail for its esses, so be heavy-handed on that send. For the full stage order, see the rap vocal chain in order; for the compressor that follows, the compression settings that pair with it.

Why do the mic and the delivery change these numbers?

Sibilance is a jet of air with a direction. Everything upstream of the plugin decides how much you have to fix.

• Distance. Three inches versus eight moves the problem more than any plugin setting does.
• Angle. Ten to fifteen degrees off-axis sends the sharpest airflow past the capsule instead of into it.
• Mic character. Bright condensers with a 5 to 10 kHz lift hand you the problem; a broadcast dynamic forgives it, which is why one sits in most rooms we cut in.
• Delivery. Pushed vocals carry more energy on every consonant, and whispered doubles are worse per decibel: the voiced part is quiet, the air is not.
• Tracking level. Peaks near full scale distort the fricatives, and distorted sibilance does not come back with gain reduction. Track peaks around −6 dBFS.

Fix any of those at the source and you need half the de-essing later. For the method side, see how to remove sibilance without dulling the vocal.

How do you know you removed esses and not words?

You listen to what was removed. A difference-monitoring mode answers this in ten seconds: you should hear esses, ts bursts and air, nothing you can read as language. If the lyric is intelligible, the crossover is too low or the threshold too deep.

Then bypass at matched output level, because louder wins any unmatched comparison, a point mastering engineer Bob Katz of Digital Domain has made for years. Check a phone speaker and earbuds too.

Building our own de-esser, we wrote probes for these failure modes:

• An ess at −9 dBFS and the same ess at −29 dBFS both received exactly 10.0 dB of gain reduction, because detection compares the high-frequency envelope to the full-band envelope in dB. A whisper and a shout get identical treatment.
• With 4.6 dB of de-essing active, 300 Hz body measured −0.00 dB. A bright vowel received 0.00 dB of gain reduction.
• A 30 ms ts burst: 8.8 dB caught, click-free. Lookahead is 3 ms, 144 samples at 48 kHz, so the gain is in place about 2 ms before the ts reaches the output.

What does too much de-essing sound like?

It has a signature you cannot unhear:

1. Esses soften into a th. That is a lisp, and the crossover is too low.
2. The front of every word softens, so the delivery drags behind the beat.
3. The top end pumps with the syllable rate, because every consonant is triggering it.
4. Doubles lose their edge and the stack collapses inward.

Backing off is a procedure, not a taste call. Cap range at 4 dB, then raise the threshold until only the two or three loudest esses move the meter, which should flicker rather than sit. More than 8 dB continuously points upstream: mic position, tracking level, or a resonance mistaken for sibilance.

Does sibilance cost you loudness on the master?

Yes, twice over, which matters if your rap masters run hot. Loudness measurement is not flat: the K-weighting curve in ITU-R BS.1770, used for the gated integrated reading in EBU R 128, adds a high-frequency shelf of about 4 dB, so a harsh top end reads louder than it feels.

Then the turn-down is automatic: Spotify's loudness normalization applies playback gain to reach its target and Apple Digital Masters asks for peak headroom that fricatives eat first. Fix esses on the vocal, where 3 dB of band-limited reduction goes unnoticed, not on the master bus.

From Our Own Line

Ours is called VEIL, and it is free. It splits with fourth-order Linkwitz-Riley filters at a crossover you set from 3 to 10 kHz (5.5 kHz default), reduces the high band only, and sums allpass-flat. SENSE maps to a relative threshold from −2 to −14 dB, capped by RANGE from 1 to 24 dB. Release is program-dependent, 20 to 200 ms base and up to three times that on a sustained shhh. Of its three LISTEN modes, Removed plays only what the plugin took.

What it is not: a broadband resonance suppressor. If your problem is a narrow ring through every word, that is resonance rather than sibilance, and oeksound's soothe2 is the tool built for it. A de-esser will not solve that, ours included. VEIL is VST3 on macOS and Windows, AU on macOS, keyless, and installs on every computer you own. Download VEIL, or see the line on the plugins page.

Frequently Asked Questions

What frequency should a de-esser be set to for rap vocals?

Between 5.5 and 8 kHz for most voices. Deep voices on a close condenser often want 4.5 to 6 kHz, brighter voices 6.5 to 8 kHz. Confirm by sweeping a boosted EQ bell from 3 to 12 kHz until the esses sting, then set the crossover there.

Should a de-esser go before or after compression?

Before. A compressor reads level, not frequency, so one loud ess makes it duck the whole word by several dB; de-essing first means the compressor responds to the voice instead of the consonants. Most rap vocals want a second light de-esser before saturation too.

How much gain reduction should a de-esser do?

3 to 6 dB on the loudest esses and nothing on the rest of the take. The meter should flicker on a few consonants per bar rather than sit engaged. Needing more than 8 dB continuously points upstream: mic distance, mic angle, or tracking level.

Do I need two de-essers on a rap vocal?

Often yes. One before the first compressor doing 3 to 6 dB, and a light one before saturation doing 1 to 3 dB, since drive generates harmonics right where fricatives already live. Add a short pass after an air shelf, and de-ess reverb sends harder.

Why does my vocal sound lispy after de-essing?

The crossover is too low, so the band is taking the part of the consonant that makes an ess read as an ess. Raise the crossover, cap the range at 4 dB, and raise the threshold until only the loudest esses trigger. Then compare at matched level.