Skip to main content
Audio-Editor
Back to Blog
AudioUpdated November 11, 20268 min read

How to Reduce Sibilance in Audio: Taming Harsh S Sounds with a De-Esser

Quick answer

Sibilance is the hiss of S, SH and Z, concentrated roughly between 4 and 9 kHz. A de-esser watches that band and turns it down only at the moments an S stands out, leaving the rest of the voice alone. On our reference voice with nine exaggerated S sounds, the Balanced (50% - Standard Podcast) preset lowered the 5–9 kHz band by 3.2 dB while the speech band moved 0.2 dB. Start there, and stop before the S turns into a lisp.

Harsh S sounds are the most common complaint about home voice recordings, and one of the easiest to make worse: a heavy-handed fix leaves the speaker sounding as if they have a lisp. This guide shows how to tell sibilance from general brightness, why it happens, how the de-esser in our editor decides when to act, and what it measured on a test file.

How to recognize sibilance

Sibilance is not the same as a bright voice. A bright voice sounds crisp all the time; a sibilant one sounds fine until an S, SH, Z or soft C arrives, and then a short, piercing hiss jumps out. It is most obvious on headphones, on words with several S sounds in a row ("she sells sea shells"), and after compression, which raises the quiet parts and the S sounds with them.

On a normal waveform you cannot see it, because the hiss is quiet compared to vowels. It shows clearly when you look only at the 5 to 9 kHz band over time, as in Figure 2: each S is a spike that rises 15 to 20 dB above the rest of the speech in that band.

Audio-Editor Online with a voice recording open on the De-Esser (Sibilance) tab
Figure 1: Our reference voice with nine exaggerated S sounds, opened on the De-Esser (Sibilance) tab. In the full-band waveform the S sounds are hard to spot.
Level of the 5 to 9 kHz band over six seconds, with nine S spikes before and slightly lower spikes after de-essing
Figure 2: Level of the 5–9 kHz band over time. Each S is a spike near −27 dBFS before (red) and about 3 to 4 dB lower after (blue); the speech between the spikes is unchanged.

Why sibilance happens

S and SH are made by pushing air through a narrow gap between the tongue and the teeth. The turbulence produces noise with most of its energy high up: roughly 4 to 6 kHz for deeper voices and 6 to 9 kHz for higher ones. Several things make it stand out in a recording:

  • Microphones with a presence peak. Many USB and condenser microphones boost 5 to 10 kHz to sound clear, which also boosts the S.
  • Speaking straight into the capsule. High frequencies are very directional. Turning the microphone slightly away from the mouth softens the S more than anything you can do afterwards.
  • Compression and loudness processing. They raise the average level, and the short S bursts come up with it.
  • Equalizer boosts for "air". A high shelf added for clarity lifts the S band too.

A static equalizer cut at 6 or 7 kHz would reduce the S, but it would also dull every other consonant and the natural breath of the voice for the whole recording. A de-esser cuts only when an S happens.

Diagram of a sidechain de-esser: a band-pass filter detects sibilance and an envelope controls how much of that band is subtracted
Figure 3: How the de-esser is built. A band-pass filter isolates the S band; an envelope follower with a 1 ms attack and 30 ms release decides how much of that band to subtract from the voice.

How the de-esser decides when to act

The de-esser runs the voice through a band-pass filter (Q of 1.5) centred on the voice profile you choose: 5.5 kHz for deeper voices, 7.2 kHz for brighter ones. It follows the level of that band with a fast envelope, 1 ms to react and 30 ms to let go, and compares it with the level of the whole signal. It acts only when two things are true: the band is above the detection threshold, and it carries more than about a third of the energy at that moment. Vowels fail the second test, so they pass untouched.

When it acts, it subtracts part of the band from the voice: output = input − (1 − gain) × band. The cut grows with how much the S dominates, up to 20 dB × the intensity. At the 50% of the Balanced preset the most it can remove is 10 dB, and typical S sounds get less, which is why the measured average is a few decibels.

How to de-ess a voice, step by step

The De-Esser page opens your file in the editor on the noise tools, with the De-Esser (Sibilance) tab already selected. It runs in your browser.

Step 1: Open the file and check the count

The card at the top, Dynamic Vocal De-Esser, shows how many sibilant peaks the analysis found; on our file it said "10 sibilant peaks detected" for nine inserted S bursts. The line below says which range will be processed. To limit the de-esser to part of the file, select it on the waveform first.

De-Esser panel with the sibilant peak count, presets, voice profiles and De-Essing Intensity slider
Figure 4: The De-Esser tab: the count of sibilant peaks, the three presets, the two voice profiles and the intensity slider.

Step 2: Pick a preset and a voice profile

  • Gentle (25% - Controlled Voice): for a voice that is mostly fine with a few sharp S sounds.
  • Balanced (50% - Standard Podcast): the default and what we measured.
  • Intense (80% - Harsh USB Mics): for bright microphones where S hurts at normal volume.

Then choose Male / Deep Voice (5.5 kHz) or Female / Bright Voice (7.2 kHz). The labels are a shortcut, not a rule: a deep female voice may sit better on 5.5 kHz and a bright male one on 7.2 kHz. Pick the one that makes the S quieter at the same intensity. De-Essing Intensity fine-tunes the amount from 0 to 100%.

Step 3: Listen and apply

Play a sentence full of S sounds on headphones at the level your audience will use. The S should stop stinging but still sound like an S. Then click Apply to Waveform. The waveform is updated and the change can be undone, so try two settings back to back if you are unsure.

Editor after applying the de-esser with the undo button enabled
Figure 5: After Apply to Waveform. The full-band waveform barely changes, because only the short S bursts in one band were lowered.

Step 4: Then compress and export

De-ess before the compressor and before normalizing. Both raise the quiet parts of the voice; if the S sounds are still too loud at that point, they rise with everything else. Export from the editor when the order is done.

Measured example

We added nine bursts of noise centered at 6.5 kHz, about the length of an S, to the reference voice. The page opened the editor on the De-Esser tab; we clicked Apply to Waveform with the default Balanced preset and the 5.5 kHz voice profile.

Energy in the 5–9 kHz sibilance band fell 3.2 dB, while the 300 Hz–3.4 kHz speech band changed +0.2 dB.

MeasuredBeforeAfter
Sibilance band 5–9 kHz26.8 dB23.6 dB
Speech band 300 Hz–3.4 kHz34.7 dB34.9 dB

Reference file (before)

Exported file (after)

Balanced is deliberately moderate, so the S sounds get softer without a lisp. For harsher recordings, the Intense preset or the 7.2 kHz profile reduce more; listen for a lisp after each change.

Download the reference file (WAV) · Same test on the tool page · Measured on November 11, 2026

The result is deliberately moderate. Balanced reduced the S band by 3.2 dB on average and the individual S peaks by about 3 to 4 dB, without touching the vowels. That is enough to take the sting out of normal sibilance. For a voice where S sounds are 10 dB too loud, Intense or the other voice profile goes further, and that is exactly where the lisp risk starts, so compare each step by ear.

Troubleshooting

What you hear or seeLikely causeWhat to change
S sounds still pierceIntensity too low, or the S sits above or below the chosen profileTry Intense, or switch voice profile
The voice sounds lispy or the S sounds like THToo much reductionUndo and use Gentle or a lower intensity
The whole voice sounds dullerHiss that is constant, not sibilanceUse background noise reduction, or a gentle equalizer cut instead
Cymbals and hi-hats in the music got quieterThey share the S bandDe-ess only the voice track, or select only the speech parts
Sibilance got worse after compressionCompression came firstDe-ess before compressing and normalizing

Limits and alternatives

A de-esser reduces sibilance; it does not fix a microphone that is harsh at every frequency or a recording full of hiss. It also cannot tell an S from a cymbal: on a mixed song it will duck both. Dedicated plugins add a split-band mode, look-ahead and frequency tracking that follows each S; this tool uses one fixed band per voice profile and reacts in real time, without looking ahead.

The biggest improvement comes before recording. Angle the microphone 15 to 30 degrees off the mouth, or point it at the chin, and keep a hand's width of distance. A foam windscreen does little for S; angle and distance do most of the work.

Frequently asked questions

What frequency is sibilance?

Roughly 4 to 9 kHz. Deeper voices put most of it around 5 to 6 kHz, brighter voices around 6 to 8 kHz. The two voice profiles centre the detector at 5.5 and 7.2 kHz.

How much reduction is right?

Enough that the S stops standing out on headphones, usually 3 to 6 dB on the S peaks. Our Balanced test gave about 3 to 4 dB. More than that tends to sound like a lisp.

Should I de-ess before or after compression?

Before. Compression raises the S along with everything else; removing the excess first keeps it under control.

Can I de-ess a full song?

You can, but cymbals and hi-hats share the band and will be ducked too. If you only have the mix, select the vocal-only passages, or use a gentle intensity.

Is the de-esser the same as a low-pass filter?

No. A low-pass filter removes the highs all the time; the de-esser acts only during an S and only on its band, so the rest of the voice keeps its clarity.

Sources and how this was tested

Tested on November 11, 2026 in Chromium 153 on Windows 11, with a production build of this site. Screenshots and measurements were made by the scripts in scripts/tool-evidence, from synthetic reference files anyone can download and repeat.

BD

Written by Bruno Dissenha

Bruno develops and maintains Audio-Editor Online. The articles document decisions, tests, and limitations observed in the current implementation.