How to reduce sibilance so listeners don't reach for the volume
- Step 1Drop the recording onto the de-esser — Add the voice file you want to make less fatiguing. MP3, WAV, FLAC, M4A, OGG, Opus, or video are accepted. One file per run.
- Step 2Find your most fatiguing passage — Scrub to the section that makes you wince on headphones — usually a run of esses. That sample is your reference for how hard to cut.
- Step 3Start low and work up — Set the Intensity (0-1) slider to
0.4first (a gentle-4.8 dB). Listen. If still fatiguing, step up by0.05-0.1and re-run. This bottom-up approach finds the smallest cut that works, which protects the voice from dulling. - Step 4Process and download — FFmpeg WebAssembly applies the notch and re-encodes to the same format you supplied. Download the smoother result.
- Step 5Listen on the playback device your audience uses — Fatigue is device-dependent. Earbuds and laptop speakers exaggerate the highs differently than studio monitors. Check on the device your audience actually uses before finalising the Intensity.
- Step 6Pair with a loudness pass if needed — Once the harsh spikes are gone you can usually raise the overall level without fatigue. Follow with loudness-normalizer to land on a comfortable, consistent loudness.
Fatigue level → Intensity to try
A restraint-first guide: start at the lowest value that fits your symptom and only go up if needed. Cut depth is exactly intensity x -12 dB at the 7-9 kHz trough.
| How fatiguing? | Intensity to try | Cut depth | Risk if you go higher |
|---|---|---|---|
| Slightly bright, occasional sting | 0.3-0.4 | -3.6 to -4.8 dB | Minimal — safe range |
| Noticeably sharp esses | 0.5-0.6 | -6 to -7.2 dB | Slight loss of air if pushed |
| Painful, ear-piercing | 0.7-0.8 | -8.4 to -9.6 dB | Voice starts to dull; A/B carefully |
| Severe, whistle-like | 0.9-1.0 | -10.8 to -12 dB | Real lisping risk — last resort only |
Symptom → is it really sibilance?
The de-esser only helps if the harshness is in its band. Use this to confirm before reaching for it, so you don't waste passes.
| What you hear | Likely band | Right tool |
|---|---|---|
| S/SH stab, ice-pick on esses | 7-9 kHz | This de-esser |
| Whistly, almost a tone on esses | 8-10 kHz | This de-esser, higher Intensity |
| Boxy, hollow, congested | 300-500 Hz | voice-eq |
| Nasal, honky | 1-2 kHz | voice-eq |
| Constant hiss / fan / AC | Broadband | ai-noise-reducer |
Cookbook
Fatigue-reduction recipes built around restraint. Each one names the Intensity to start at and the symptom it targets.
Audiobook narration that tires the ear over an hour
Long-form listening punishes even mild sibilance. A gentle cut keeps the narration comfortable for the full runtime without sounding processed.
Content : audiobook chapter, 58 min WAV Goal : comfortable for a full-length listen Control : Intensity = 0.45 (-5.4 dB notch) Result : esses softened just enough; air preserved Why low : over-cutting dulls narration over a long session
Lecture recording with a bright lapel mic
Educational content is often listened to at higher volume to catch detail, which makes sibilance worse. A medium cut helps without losing the clarity students need.
Content : 90 min lecture, lapel mic, MP3 Problem: esses sharp at study volume Control : Intensity = 0.6 (-7.2 dB) Result : clear and far less fatiguing for long study
Working up from the bottom to avoid over-cutting
The restraint-first method in practice: three quick passes to find the floor rather than guessing high and dulling the voice.
Pass 1: Intensity 0.4 -> still a little sharp Pass 2: Intensity 0.5 -> comfortable Pass 3: not needed Keep : 0.5 (-6 dB) — the smallest cut that worked
Confirming it's sibilance before you cut
If the harshness isn't in the band, no Intensity helps. A quick check saves wasted passes.
Test : set Intensity 1.0, process a 10s clip
If harshness gone -> it was sibilance, dial back down
If still harsh -> not in 7-9 kHz band; use voice-eq
for the real culprit insteadDe-ess then raise the level comfortably
Once spikes are tamed, the ceiling for comfortable loudness rises. De-ess first, then normalize.
Step 1: de-esser, Intensity 0.55
Step 2: loudness-normalizer -> -16 LUFS
Result : louder overall yet less fatiguing, because the
peaks that hurt are no longer thereEdge cases and what actually happens
Cutting too deep makes the voice lispy
ExpectedThe most common over-correction. Toward 1.0 (-12 dB) the notch removes enough high-frequency energy that 's' starts to sound like 'th' and the voice dulls. Restraint is the cure — use the smallest Intensity that removes the sting.
Harshness persists at maximum Intensity
Out of bandIf 1.0 doesn't fix it, the harshness isn't sibilance in the 7-9 kHz band — it's elsewhere in the spectrum. The de-esser cannot reach it. Diagnose with the symptom table and switch to voice-eq or ai-noise-reducer.
Breathy / intimate narration loses its air
ExpectedIntentional breathiness lives in the same band the de-esser cuts. Heavy Intensity strips that character. For intimate narration, keep Intensity at 0.3-0.4 so you tame spikes without flattening the air.
Free plan blocks processing
Pro requiredThe de-esser is Pro-only. On Free it will not run. Upgrade to Pro for files up to 200 MB / 120 minutes.
Recording exceeds the duration cap
RejectedPro caps each file at 120 minutes. A 2.5-hour seminar is rejected on Pro for duration. Trim with audio-trimmer, split with audio-splitter, or upgrade to Pro-media / Developer (unlimited duration).
Output format differs from what you expected
By designThe de-esser re-encodes to the same format you supplied — there is no output-format choice. Convert afterwards with the matching converter if you need a different format.
No batch — many lectures to process
One file at a timeProcess recordings one by one. Because the curve is fixed, applying the same Intensity to each gives uniform fatigue reduction across a course or series.
Fatigue only appears on cheap earbuds
Device-dependentSome earbuds boost the highs, exaggerating sibilance that sounds fine on monitors. The de-esser helps, but check the result on the device your audience uses — a setting that sounds over-cut on monitors may be right for earbud listeners.
Music or sound effects in the file
Affects highsThe notch sits on top of any high-frequency content, including cymbals and bright effects. On mixed content it dulls those too. De-ess the voice before mixing in music if you want the bed unaffected.
Frequently asked questions
What exactly causes listener fatigue, and how does this help?
Sharp s/sh energy in the 5-10 kHz band stabs the ear and forces listeners to ride the volume. This de-esser cuts a fixed notch deepest at 7-9 kHz, reducing those spikes so the voice stays at a comfortable, even loudness without the stabs that cause fatigue.
How do I reduce sibilance without making the voice dull?
Use restraint. Start the Intensity low (0.4) and increase only until the sting is gone. The cut depth is intensity x -12 dB, so small numbers are gentle. Stop at the smallest value that works — over-cutting (toward 1.0) is what dulls the voice and causes lisping.
Is this a dynamic de-esser?
No. It applies a fixed EQ notch (firequalizer) across 4500-12000 Hz to the whole file rather than ducking only on sibilant peaks. The benefit is predictability and consistency; the trade-off is that the cut is always on rather than level-triggered.
Why start the Intensity low instead of high?
Because the smallest effective cut preserves the most of the voice. Starting low and stepping up finds your floor, where listener fatigue is gone but air and clarity remain. Starting high risks dulling the voice before you realise less would have worked.
Does the recording get uploaded?
No. Processing runs on FFmpeg WebAssembly in your browser. Interview, lecture, and audiobook recordings are filtered entirely on your machine and never sent to a server.
What if the harshness isn't sibilance?
Then the de-esser won't help — it only touches 4500-12000 Hz. Use the symptom table above: boxiness and nasality (below the band) are jobs for voice-eq; constant hiss is for ai-noise-reducer.
Can the same setting work across a whole course of lectures?
Yes. The fixed curve means the same Intensity produces the same reduction every time. Calibrate once on a representative recording and reuse the value for uniform results across the whole series.
What's the maximum reduction this tool can apply?
Intensity 1.0 gives a -12 dB notch at the 7-9 kHz trough. That's the deepest cut available. If 1.0 still doesn't fix the harshness, the problem isn't sibilance in this band and a different tool is needed.
Will de-essing let me raise my overall volume?
Often, yes. The harsh peaks are what cap comfortable loudness; once they're tamed you can usually push the overall level higher without fatigue. De-ess first, then run loudness-normalizer.
How long a recording can I process?
Pro handles up to 120 minutes and 200 MB per file. For longer recordings, split with audio-splitter or upgrade to Pro-media / Developer, which lift the duration cap and allow files up to 100 GB.
Is this available on the Free plan?
No — the de-esser requires Pro or higher. Upgrade to Pro to process files up to 200 MB / 120 minutes; Pro-media and Developer add unlimited duration.
What output format do I get back?
The same format you dropped in, re-encoded with the matching codec. There is no format selector for this tool. Convert afterwards if you need a different format.
Privacy first
Every JAD Audio tool runs entirely in your browser via FFmpeg (WebAssembly) and RNNoise. Your audio files never leave your device — verified by zero outbound network requests during processing.