Mic processing
How to tame harsh s sounds with a de-esser
Every sharp s your mic exaggerates lives in one narrow band of hiss, roughly 5 to 8 kHz. A de-esser watches exactly that band and turns it down only in the instant it flares. The setup is three steps, and the defaults do most of the work.
This guide uses the de-esser in patchd, a free node on every channel. If you want the theory first, what a de-esser is covers the concept in depth; this page is the setup: add the node in the right slot, trust the defaults, and spend two minutes tuning by ear.
First, see where the harshness lives
Sibilance is the burst of high-frequency hiss your mouth makes on s, sh, ch, and z sounds. The tone of your voice sits mostly below 4 kHz, but those bursts concentrate their energy higher, around 5 to 8 kHz for most voices, and close miking plus the presence boost built into most vocal mics sharpens every one of them into a little dart. That localization is exactly why a de-esser works: the problem occupies a narrow band, so the fix can watch that band and ignore everything else. The patchd de-esser draws this on screen, and its default detection band lands right on top of the trouble zone.
Step 1: add the de-esser after your compressor
Open your mic channel's effects rack and pick De-esser from the + Add Node menu; it sits in the dynamics group and it is free on every channel. Where you drop it matters more than usual. A compressor raises your average level, and its makeup gain lifts everything, including the sibilance that was already the sharpest thing in the signal. A voice that barely needed de-essing before the compressor clearly needs it after.
So the de-esser goes directly after the compressor, where it cleans up exactly what compression emphasized. If it landed in the wrong slot, dragging a node reorders it. The broader logic of who feeds whom is covered in chain order.
Step 2: start from the defaults
The node ships tuned for a typical voice, so the fastest setup is to trust the starting values and only move what the next step tells you to. Four controls define the whole beginner tier:
- Threshold (default -28 dB). The line where de-essing engages, measured inside the detection band only. Lower it to catch more sibilance, raise it to be more selective.
- Range (default 8 dB). The maximum reduction the node may apply, from 0 to 20 dB. This is the guardrail against the lispy, over-de-essed sound: 6 to 10 dB reads natural, above 12 starts to dull the voice.
- Frequency (default 6500 Hz). The center of the detection band, adjustable from 2000 to 12000 Hz. Most voices hiss loudest near the default.
- Bandwidth (default 1.5x). How wide the band is around that center, from a surgical 0.5x to a very wide 5x.
One more default worth knowing before you touch anything: the Mode toggle ships as Split-band, which compresses only the sibilant frequencies and leaves the body of your voice untouched even mid-s. Broadband, the alternative, ducks the whole signal when sibilance hits, which is simpler but can dip your low end on s-heavy phrases. Stay on Split-band unless you have a reason not to.
Step 3: tune it by ear
The node's meters make this a two-minute job. The DETECT bar shows the level inside the band with your threshold drawn across it, and the GR bar shows how much reduction is actually happening.
- Speak a few s-heavy sentences at your normal distance and volume. "Sixty silver spoons on a seaside session" earns its keep here.
- Watch the DETECT bar: your s bursts should spike above the THR line while normal speech stays below it. Lower the threshold a few dB at a time until every harsh s makes the GR bar tick.
- Press DELTA and talk. You are now hearing only what the de-esser removes, and it should sound like pure hiss. If whole words leak through, the threshold went too low; back it up.
- Stop before the lisp. If your s sounds start disappearing rather than softening, raise the threshold, and let the Range cap hold the reduction to a natural 8 dB no matter how hard a burst hits.
The finish line is anticlimactic on purpose: your voice should sound exactly like you, just without the darts. A de-esser you can hear working is a de-esser set too hard.
If it still sounds harsh, move the band
Not every voice hisses at 6500 Hz. Male voices tend to concentrate sibilance around 5 to 7 kHz, female voices around 6 to 9 kHz, and your mic's own brightness shifts it further. The LISTEN button on the node solos the detection band itself, what the detector hears, so sweep the Frequency slider while listening until the hiss is at its loudest, then let go. If the harshness spans a wider stretch, raise Bandwidth; if it is one specific whistle, narrow it. The two controls define the band edges exactly:
f_low = fc / sqrt(bw), f_high = fc * sqrt(bw)At the defaults, sqrt(1.5) is about 1.22, so the detector listens from about 5307 Hz up to about 7961 Hz. The multiplier splits evenly across both sides of the center, which is why the shaded band on the display sits symmetric on the log axis.
The Intermediate group holds the two remaining character controls. If tuned sibilance still cuts through, Ratio (default 6:1) can climb toward 10:1 for a more surgical clamp, and the Range cap still limits the total reduction regardless. And if your voice sounds thinner on s-heavy phrases after you experimented with Broadband mode, that dip in body is the mode, not the settings: switch back to Split-band.
When the de-esser needs EQ help
A de-esser fixes bursts. If your voice sounds harsh on every word, vowels included, that is a tone problem, and the right tool is EQ: ease the presence boost around 3 to 6 kHz, or add a gentle high-shelf cut. What you should not do is EQ away the 5 to 8 kHz region wholesale to kill sibilance. A static cut dulls the entire voice all the time, while a de-esser dips only during the s and leaves every other syllable bright. The two compose well: a small shelf cut tames a genuinely bright mic, and the de-esser handles whatever bursts remain. If you are still diagnosing whether sibilance is even your problem, the symptom checklist in fixing harsh s sounds walks the full lineup of causes.
In patchd the de-esser is free on every channel, and the display, meters, and monitor buttons pictured here are the node's real surface, so what you tune is what every destination hears. patchd is in development now. Join the waitlist and we will tell you the moment it is ready to install.