Sibilance is the sharp, sustained energy that makes S, SH and Z sounds jump out of an otherwise balanced voice. The wrong repair is easy to hear: the consonants collapse, the speaker starts to sound as if they have a lisp, and the whole recording loses its air. The safer route is to work out the shape of the problem first, then use the narrowest treatment that fixes it.
Confirm that the problem is actually sibilance
A bright consonant doesn't automatically call for de-essing. Listen to how long it lasts and where the energy sits before you pick a tool.
| What you hear | What it usually is | Better first move |
|---|---|---|
| A long, bright hiss on S, SH or Z | Sibilance | Manual level repair, dynamic EQ or a de-esser |
| A low thump on P or B | Plosive | Local level and low-frequency repair; see the plosive removal workflow |
| A tiny isolated tick | Mouth click or edit click | Short event repair; compare with the mouth-click guide |
| Every word feels edgy, not just consonants | Broad tonal imbalance, clipping or microphone coloration | Check the recording, then consider broader EQ or restoration |
Find the harsh band by ear first
Start with the most obvious S sound in the file. Loop a short phrase, but keep enough of the vowel before and after it to hear whether the voice stays natural. A spectrum graph can show a concentrated rise in the upper mids or treble, but let it confirm what you hear. It shouldn't pick the band for you.
A useful search area often sits somewhere around 4–10 kHz, and that's only a starting region. The speaker, microphone angle, room, codec and earlier processing can all move the painful band. Sweep a narrow monitoring band until the hiss is unmistakable, then widen it just enough to catch the full consonant. If the vowels start to sound papery, the band is too wide or too low.
Repair isolated syllables before reaching for a processor
When only a handful of consonants are distracting, manual repair usually sounds cleaner than running a de-esser across the whole file.
- Preserve the source. Save a new version before destructive edits.
- Select the sibilant portion. Include the sustained hiss and leave the rest of the word alone.
- Reduce level gently. Use gain reduction or a short envelope and listen to the transition into the following vowel.
- Shape only if level repair is not enough. Apply a small local EQ cut to the selected event instead of pulling treble out of every sentence.
- Check the boundary. Add a tiny fade if the edit creates a click, then replay the complete phrase.
This method takes longer when a recording has dozens of similar events. What you get in return is control: a soft S can stay untouched while one aggressive consonant gets a precise edit.
Use Dynamic EQ for repeated sibilance
Sound Forge 2026 documentation describes Dynamic EQ as a Suite feature. If it's in your installed edition, it suits repeated sibilance well, because the selected band only changes when its level triggers the processor. Confirm your license and available effects before you build the workflow around it.
- Create a bell band around the harshest part of the S sound.
- Set the band to reduce level when that region becomes dominant.
- Lower the threshold until strong sibilants trigger reliably, while vowels and cymbals pass without obvious movement.
- Use a quick attack to catch the front of the consonant. Set the release so the reduction recovers before the next vowel without fluttering.
- Start with a small amount of reduction. Increase it only while the word keeps getting easier to hear, and back off as soon as the consonant loses definition.
In the documented Dynamic EQ, that level-dependent behavior comes from Band Dynamics. Switch it on for the selected band, then set Threshold, Ratio, Attack and Release as you would on a compressor: a ratio above 1 pulls the band down once it crosses the threshold. It's still an EQ band, though, and the documentation lists no detector-listen control like the ones dedicated de-essers offer.
Do not use static EQ as an automatic de-esser
Static EQ applies the same cut whenever audio passes through. That's useful when the entire recording is too bright. When the problem only shows up on certain consonants, it's wasteful, because it also strips presence from vowels, breaths and room tone.
If the voice is harsh throughout, use the Sound Forge equalizer guide to fix the broader balance first, then reassess the S sounds that are left. A smaller de-essing stage after tonal correction often sounds more natural than asking one processor to solve both problems.
When a dedicated VST3 de-esser is the better choice
Current Sound Forge documentation covers VST3 plug-ins. A dedicated de-esser earns its place when you need to audition the detector signal, switch between wide-band and split-band processing, or set the detection range more precisely. Check the plug-in format, scan path and license on your own system before you commit a project to it.
| Mode | What moves when sibilance is detected | Where it tends to help | Main risk |
|---|---|---|---|
| Wide-band | The whole signal turns down briefly | Exposed speech and vocals where the level dip is unobtrusive | Pumping or a small audible volume hole |
| Split-band | Mainly the high-frequency region turns down | Dense mixes or voices that must keep their body stable | A dull or phasey top end if the band is too broad |
Use the detector-listen function if the plug-in has one. You should hear mostly the offending hiss. If whole syllables, vowels or musical detail dominate the detector signal, narrow or move the range before you add more reduction.
Prevent lisping, dullness and pumping
- Lisping: reduce the range or depth until T, S and SH remain distinct.
- Dull vowels: move the band higher, narrow it, or use split-band processing.
- Pumping: reduce the wide-band amount or shorten the recovery without making it chatter.
- Missed consonants: adjust the detector range before lowering the threshold further.
- Inconsistent results: handle the few outliers with automation or manual edits instead of forcing one setting to catch every event.
Don't judge a de-esser by soloing the S sounds. You aren't trying to erase them. You want them to sit inside the sentence.
Run the final check in context
Play the processed voice back at a normal listening level and compare it with the bypassed version at similar loudness, on headphones and on speakers. Then listen to a quiet S, a strong S, an SH sound and a phrase with no sibilance at all. If that last phrase changes noticeably, the processor is working too often.
For voice-over work, check the repair before the final limiter or loudness stage, since later compression can push missed consonants forward again. The voice-over workflow covers the broader processing order, and the export QA checklist gives you one last pass before delivery.