Audio foundations
De-esser vs EQ for sibilance
Both can soften a harsh s. Only one of them stops when the s does. An EQ cut is subtracted from every sound you make, all the time; a de-esser is a compressor that watches the sibilant band and clamps only in the moments an s is actually happening. Static versus dynamic is the whole decision.
If either tool is new to you, start with what a de-esser is or what an EQ is. This page is the decision between them: why the obvious EQ move quietly charges your whole voice for a problem three letters cause, what the de-esser's numbers actually do, and the honest cases where EQ is still the right answer.
Sibilance is an event, not a tone
Say "this is us" out loud, the phrase drawn in the chart further down. Each s is a burst of high-frequency hiss lasting maybe 50 to 150 ms, with its energy concentrated roughly between 2 and 12 kHz and, for most voices, loudest somewhere near 6.5 kHz. Everything between the esses is a different animal: vowels, m and n sounds, tiny silences. In that same 6.5 kHz neighborhood those sounds carry consonant detail and the airy top of your voice, and they sit far quieter in the band than an s does.
That makes sibilance a timing problem wearing a frequency costume. The offending band is only offensive for the fraction of a second an s occupies; the rest of the time it is doing useful work. A fix that runs one hundred percent of the time to solve a problem present a few percent of the time charges rent on the other ninety-plus. Hold that thought against the two tools.
The static fix: a cut that never sleeps
The instinctive move is EQ, because the problem sounds like a frequency: find where the s lives and pull it down. Drag a peak band up to 6.5 kHz, cut 6 dB, tighten the Q to around 2 so it does not swallow the presence range below it. And it works, partially. Every s gets 6 dB quieter.
So does everything else. An EQ band is a fixed reshaping of the frequency response; nothing on the card can know whether an s is happening. That 6 dB comes off the air on every vowel, the crispness of every t and k, the sheen your mic's presence peak was providing, on every word and straight through the silences. Meanwhile the s itself was maybe 12 or 14 dB too hot, so after paying 6 dB across your entire voice it is still the sharpest thing in the signal. Cut deeper and the s finally behaves at about the moment your voice goes dull and slightly lisped. The static tool cannot win this one, because it has to charge every sound the same price.
The dynamic fix: a clamp that exists only during the s
A de-esser is a compressor whose detector listens only to the sibilant band, and on the patchd de-esser the numbers are concrete. It watches a band centered at 6500 Hz with a bandwidth of 1.5x, roughly 4.3 to 9.8 kHz. Normal speech in that band idles far below the -28 dB threshold, so the node does nothing, which is to say it costs nothing. When an s spikes the band across the threshold, the node compresses it at 6:1, never by more than the 8 dB Range cap, arriving in about a millisecond and letting go over a 60 ms release. The timing is sized to the event: a clamp that lives and dies inside a single consonant.
Read the bottom lane. The dashed line is the EQ cut: minus six, everywhere, forever. The green line is the de-esser: flat zero between esses, the full 8 dB on the two hot bursts, and only about 5 dB on the milder middle one, because a compressor's reduction is proportional and a gentler s asks for a gentler clamp. A static cut deep enough to fix the worst s would over-punish the mild ones; the dynamic tool sizes every correction to the moment it happens in.
When EQ is still the right tool
None of this makes EQ the wrong tool. It makes it the wrong tool for intermittent problems. When the complaint is constant, a constant fix is exactly correct: a voice that sounds harsh on every word, not just on esses, has a tonal balance problem, and that is EQ territory, a gentle wide cut where the harshness lives, judged over whole sentences. Boxiness around 250 to 500 Hz, rumble below 80 Hz, a dull top that wants a high shelf: all static problems, all EQ jobs, all walked through in how to EQ your microphone.
The two tools also pair better than they compete. The classic trap is a presence or air boost that makes a voice clearer and its esses sharper at the same time; most people back the boost off and lose the clarity they wanted. Owning both tools breaks the trade: let the EQ set the bright tone on purpose, and let the de-esser police that band's worst moments. The warning runs the other way too. A de-esser asked to fix a voice that is harsh all the time will sit clamped most of the time, which is just a worse EQ with pumping. If it is always broken, it is tone. If it breaks on s sounds, it is sibilance.
Two formulas, one difference
The whole page compresses into where each tool's gain comes from. An EQ band is an RBJ biquad, and its response is a function of frequency alone:
gain_EQ(f, t) = G(f)Evaluate the -6 dB peak at 6.5 kHz and G(6500) is -6 whether you are mid-s, mid-vowel, or silent. The filter is deaf by design; nothing the signal does can change it.
The de-esser's gain is a function of what the band is doing right now:
reduction_dB(t) = min(Range, (1 - 1/R) * (bandDb(t) - T))At the defaults (threshold -28, ratio 6:1, range 8), a burst reading -16 dB in the band is 12 dB over, asks for 10 dB of reduction, and Range caps it at 8. Between esses the band reads around -45, the bracket goes negative, and the reduction is zero.
Same units, same direction, one substitution: G(f) is frozen and bandDb(t) is alive. The clamp even leaves gracefully, decaying on the 60 ms release the way every compressor does. That one substitution is the entire difference between reshaping a voice and supervising one.
Which one, then?
| Static EQ cut | De-esser | |
|---|---|---|
| What it is | A fixed dip in the frequency response | A compressor watching one band |
| When it acts | Always, on every sound | Only while the band crosses -28 dB |
| What it costs | Air and clarity on every non-s sound | Nothing between esses |
| Sized to the problem? | One depth fits all | Proportional, capped by Range |
| Pick it when | The harshness is on every word | The harshness comes and goes with the s |
For sibilance itself, the de-esser wins, because sibilance is a dynamic problem and the de-esser is the dynamic tool. Reach for the EQ when the complaint survives a whole sentence. In patchd both are free nodes on every channel, living in the same effects rack: put the EQ early for cleanup and tone, put the de-esser after the compressor, since compression pushes sibilance up along with everything else, and let each tool do the job it is shaped for. patchd is in development now. Join the waitlist and we will tell you the moment it is ready to install.