How to remove harsh sibilance from voice recordings
- Step 1Open the de-esser and drop your recording — Drop a voice file onto the de-esser. It accepts MP3, WAV, FLAC, M4A, OGG, Opus, or a video file (the audio track is extracted). Processing is one file at a time — this tool does not batch.
- Step 2Listen for where the harshness lives — Before touching Intensity, play the raw file on headphones. The de-esser cuts a fixed band centred on
7000-9000 Hz. If your harshness sits there (most voices do), this tool fits. If the harshness is lower (a boxy 300-500 Hz) it is not sibilance — use voice-eq instead. - Step 3Set the Intensity — Drag the single Intensity (0-1) control. It defaults to
0.6(a-7.2 dBnotch). Step size is0.05. Start there; only push higher if esses are still painful. The relationship is exact: the cut depth isintensity x -12 dB. - Step 4Process and download — Click process. FFmpeg WebAssembly applies the
firequalizernotch and re-encodes to the same format your file came in as. The result downloads straight to your machine. - Step 5A/B the result against the original — Compare before/after on headphones. The goal is smooth, not silent — esses should still be audible. If the voice now sounds dull or lispy, you over-did it: lower the Intensity and re-run. The fixed curve means a lower value is the only lever.
- Step 6Chain it into a finishing pass if needed — De-essing changes the high-frequency balance, so do it before loudness. Run loudness-normalizer afterwards for a consistent LUFS target, or use podcast-master for a full denoise-EQ-level-normalize chain in one click.
Intensity setting → actual EQ cut
The Intensity control maps linearly to the notch depth in the firequalizer curve: cut = intensity x -12 dB. The band is flat at 4500 Hz, sits at the cut depth from 7000-9000 Hz, and returns to flat by 12000 Hz. Values below are exact, not approximate.
| Intensity | Cut at 7-9 kHz | Character | Good for |
|---|---|---|---|
0.3 | -3.6 dB | Barely-there shave | Lightly bright voices that only spike on the occasional word |
0.5 | -6 dB | Gentle, transparent | Most voiceover where you want to keep airiness |
0.6 (default) | -7.2 dB | Balanced | A typical condenser-mic vocal — the safe starting point |
0.8 | -9.6 dB | Strong | Persistent sibilance from a very bright mic or boosted top end |
1.0 | -12 dB | Maximum (risk of dulling) | Severe esses only — listen for lisping before committing |
What this tool is, and what it is not
The de-esser is a fixed-curve EQ notch, not a studio dynamic de-esser. Knowing the difference prevents wasted time hunting for controls that do not exist.
| Capability | This tool | Where it lives instead |
|---|---|---|
| Reduce 7-9 kHz sibilance | Yes — fixed firequalizer notch, depth set by Intensity | This tool |
| Level-triggered / dynamic reduction | No — the cut is always on, applied to the whole file | Not available; use a DAW de-esser plugin for true dynamic behaviour |
| Adjustable centre frequency | No — the band is hard-coded to 4500-12000 Hz | Not available |
| Threshold / Mild-Medium-Aggressive presets | No — only the continuous Intensity (0-1) control | Not available |
| Broad tonal shaping (high-pass, presence, warmth) | No | voice-eq |
| Even out loud/quiet words | No | speech-leveler |
Tier limits for the de-esser
The de-esser requires Pro or higher — it is not available on the Free plan. Limits are per file. Duration is a separate cap from file size.
| Plan | Max file size | Max duration | Files per run |
|---|---|---|---|
| Free | Not available (Pro-only tool) | — | — |
| Pro | 200 MB | 120 min | 1 (no batch) |
| Pro-media | 100 GB | Unlimited | 1 (no batch) |
| Developer | 100 GB | Unlimited | 1 (no batch) |
Cookbook
Concrete settings for common voice-recording problems. Intensity is the only control, so each recipe is just the right value plus what to listen for.
A bright voice that spikes on every 'S'
The classic case: a clear, forward voice where individual S words jump out and stab. The default Intensity handles most of these without dulling the tone.
Symptom : 's' words ('sister', 'success') stab on headphones
Control : Intensity = 0.6 (default)
Result : -7.2 dB notch at 7-9 kHz
Verdict : Esses sit back into the voice; airiness preserved
If still harsh: nudge to 0.7 (-8.4 dB), re-run, A/BOver-de-essed and now lispy — back it off
Pushing Intensity to 1.0 can replace a crisp 's' with a soft 'th'. Because the curve is fixed, the fix is simply a lower value — there is no attack/release to blame.
Symptom : 'sun' now sounds like 'thun', voice feels dull Was : Intensity = 1.0 (-12 dB notch) Fix : Intensity = 0.5 (-6 dB notch) Result : Sibilance tamed without the lisp Rule : smallest cut that fixes the problem wins
WAV in, WAV out — keep a lossless master
The tool re-encodes to the same format you dropped in. A 24-bit recording session WAV stays WAV (re-encoded to 16-bit PCM), so you can de-ess your master before any lossy export.
Input : interview-master.wav (44.1 kHz)
Control: Intensity = 0.6
Output : interview-master-deessed.wav (pcm_s16le)
Note : output format always matches input; there is no
format dropdown for the de-esser in the UITwo voices on one mono recording
When two speakers share a track but only one is sibilant, the fixed full-file notch still helps the harsh voice while leaving the other essentially untouched (its energy in 7-9 kHz is already low). Set Intensity for the worst offender.
Input : two-host-segment.mp3
Problem: Host A is sibilant, Host B is not
Control: Intensity = 0.55 (tuned to Host A)
Result : Host A smoothed; Host B barely affected because
his speech has little 7-9 kHz energy to cutDe-ess first, then normalize loudness
De-essing removes high-frequency energy, which slightly lowers the integrated loudness. Always de-ess before you hit a final LUFS target so the level lands where you expect.
Step 1: de-esser, Intensity = 0.6
Step 2: loudness-normalizer -> -16 LUFS (Apple Podcasts)
Why : doing loudness last means the de-ess cut is already
baked in before the gain stage measures the fileEdge cases and what actually happens
Harshness is not actually sibilance
By designThe de-esser only cuts 4500-12000 Hz. If your recording sounds harsh in the lower mids (boxy 300-500 Hz, or honky 1-2 kHz), no Intensity setting will fix it because that energy is below the band. Reach for voice-eq for broad tonal shaping instead.
Intensity pushed to 1.0 makes the voice lispy
ExpectedA -12 dB notch removes enough high-frequency energy that crisp esses can soften into a 'th' sound. This is the over-de-essing failure mode. It is not a bug — lower the Intensity. The curve is fixed, so a smaller value is the only correction.
Output is the same format as the input, not MP3
By designThe de-esser UI has no output-format selector. It re-encodes to whatever format you dropped in (FLAC stays FLAC, WAV stays WAV). If you specifically need to change container or codec, do that after de-essing with a converter such as wav-to-mp3.
Free plan blocks the tool
Pro requiredThe de-esser is a Pro-tier feature (minTier: pro). On the Free plan it will not process. Upgrade to Pro for 200 MB / 120-minute files, or to Pro-media / Developer for 100 GB and unlimited duration.
File exceeds your tier's size or duration cap
RejectedPro caps each file at 200 MB and 120 minutes. A 3-hour raw recording will be rejected on Pro for the duration cap even if it is under 200 MB. Trim it first with audio-trimmer, or upgrade to Pro-media / Developer for unlimited duration.
No batch processing
One file at a timeThe de-esser processes a single file per run (acceptsMultiple: false). To de-ess a series of episodes, run them one after another with the same Intensity — because the curve is fixed, identical settings give identical treatment, so consistency across files is automatic.
Music with cymbals or hi-hats
Supported (with care)The 7-9 kHz notch also sits on top of cymbal and hi-hat energy. On a full music mix this will dull the top end, not just tame vocal sibilance, because the tool applies the cut to the whole file rather than ducking on a side-chain. It works best on isolated voice. For a music mix, use the gentlest Intensity that helps.
Whispery / breathy recording loses air
ExpectedBreathy ASMR-style voices carry a lot of intentional high-frequency air in the same band the de-esser cuts. Heavy Intensity will remove that character. Keep it low (0.3-0.5) so you tame the spikes without flattening the breathiness.
Already-EQ'd file with a top-end boost
SupportedIf you boosted highs for clarity and that made esses worse, the de-esser counteracts it cleanly in the 7-9 kHz band. Order matters: apply your presence boost first, then de-ess, so the notch lands on the boosted material.
Very low sample-rate source (8 kHz telephone)
Limited effectAn 8 kHz-sampled phone recording has a Nyquist limit of 4 kHz, so there is little to no energy in the 7-9 kHz target band. De-essing such a file has minimal audible effect — the sibilance you hear there is below the band the tool touches.
Frequently asked questions
Is this a real dynamic de-esser like in a DAW?
No, and the page is honest about that. A studio de-esser is a level-triggered dynamic compressor that only ducks the band when sibilant energy spikes. This tool applies a fixed firequalizer EQ notch across 4500-12000 Hz (deepest at 7-9 kHz) to the whole file. The trade-off: it is fully predictable and identical every run, but it is always-on rather than triggered. For most spoken-voice recordings the fixed notch is enough; for surgical dynamic de-essing on a music vocal, a DAW plugin is the right tool.
Can I change the target frequency?
No. The band is hard-coded in the filter: flat at 4500 Hz, full cut from 7000-9000 Hz, flat again by 12000 Hz. There is no frequency control in the UI. This band covers where the vast majority of vocal sibilance lives, but if your harshness sits outside it the de-esser is the wrong tool — use voice-eq for broad shaping.
What does the Intensity control actually do?
It sets the depth of the EQ notch. The cut in decibels equals intensity x -12. So 0.6 (the default) is a -7.2 dB cut, 0.5 is -6 dB, and 1.0 is the maximum -12 dB. The slider step is 0.05. That single value is the only de-essing parameter — there is no separate threshold, ratio, attack, or release.
Why is there no Mild / Medium / Aggressive preset?
Because the continuous Intensity control already covers that range and is more flexible. Mild is roughly 0.4, Medium is the 0.6 default, and Aggressive is 0.8-1.0. There is no separate preset dropdown in the de-esser UI — set the number directly.
Will the file be uploaded anywhere?
No. The de-esser runs on FFmpeg compiled to WebAssembly inside your browser. Your recording is decoded, the firequalizer notch is applied, and the result is re-encoded entirely on your own device. The audio never leaves your machine and is never sent to a JAD server.
What formats can I drop in, and what comes out?
In: MP3, WAV, FLAC, M4A, OGG, Opus, or a video file (the audio track is used). Out: the same format you dropped in, re-encoded with the matching codec (libmp3lame for MP3, pcm_s16le for WAV, FLAC, AAC for M4A). There is no output-format selector for this tool — to convert, use a dedicated converter afterwards.
How much reduction is too much?
Trust your ears, not a number. If esses turn into 'th' sounds or the voice goes dull, you have over-de-essed — drop the Intensity. As a starting heuristic, the default 0.6 (-7.2 dB) is right for most condenser-mic voices; anything above 0.8 risks audible dulling on a normal voice.
Can I de-ess a whole podcast episode in one go?
Yes, within the tier caps. On Pro you can process a single file up to 200 MB and 120 minutes. A typical hour-long episode is well inside that. For longer raw recordings, upgrade to Pro-media or Developer (unlimited duration) or trim first with audio-trimmer.
Does de-essing change the loudness of my file?
Slightly. Removing energy in the 7-9 kHz band lowers the integrated loudness a little. That is why you should de-ess before any final loudness pass. Run loudness-normalizer afterwards so your LUFS target is measured on the already-de-essed audio.
It is on the Free plan options — why won't it run?
The de-esser is gated to Pro and above (minTier: pro). It will appear but not process on Free. Pro unlocks 200 MB / 120-minute files; Pro-media and Developer raise that to 100 GB and unlimited duration.
Can I batch a folder of clips?
Not in a single run — the de-esser takes one file at a time. The upside is consistency: because the curve is fixed, applying the same Intensity to each clip in turn gives identical treatment, so a batch processed one-by-one sounds uniform.
What if my recording is harsh but not from S sounds?
Then de-essing won't be the fix. The tool only cuts 4500-12000 Hz. Boxiness (300-500 Hz), nasality (1-2 kHz), or muddiness all live below that band — shape those with voice-eq. Background hiss or hum is better handled by ai-noise-reducer.
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.