> ## Content Index > Fetch the complete content index at: https://soundforgepro.com/llms.txt > Use this file to discover other available public pages before exploring further. # iZotope RX De-ess: Spectral vs Classic Settings for Voice - URL: https://soundforgepro.com/izotope-rx-de-ess/ - Published: 2026-10-26T07:41:00.000Z - Updated: 2026-10-26T19:01:46.000Z - Description: How to set iZotope RX 12 De-ess for dialogue and narration: when Spectral beats Classic, the order to set Cutoff, Threshold and Speed, how to hear what you remove, and what RX Elements users get in Repair Assistant. - Author: Erick Finn - Tags: iZotope RX, Voice Recording **For speech, start iZotope RX De-ess in Spectral mode. Set Cutoff freq so the S sits above it and the vowel body below, leave Threshold in its default Relative mode (Absolute unchecked) and lower it until the worst esses drop without the speaker sounding lispy. Pick Fast if the top end pumps and Slow if consonant attacks go soft. Then turn on Listen (the ear icon) and Preview to hear what you are removing, A/B with Bypass, keep alternatives in Compare and Render.** The De-ess module and plug-in come with RX 12 Standard and Advanced. Elements has no De-ess plug-in, but the Repair Assistant plug-in, included in every RX 12 edition, has a simplified De-Ess section in its Voice mode. My default for dialogue, podcasts and narration is Spectral with a Relative threshold. It leaves the vowel body alone, and it keeps following a speaker whose level moves from line to line. I'd switch to Classic only when the whole consonant jumps out as one loud event and a short dip across the word sounds more natural than a trimmed top end. I'd also keep each pass light: two gentle passes, or a manual fix on the few worst esses plus one gentle pass, usually beat one heavy setting. De-ess is one of the voice-repair tools covered in the [iZotope RX guides](https://soundforgepro.com/izotope-rx/), and everything below follows the [RX 12 De-ess manual page](https://docs.izotope.com/rx12/en/de-ess.html). ## Where De-ess lives in RX 12, and what Elements gets instead The RX 12 manual tags De-ess as a Standard and Advanced feature that works in two places: as a module inside the RX Audio Editor and as a plug-in in your DAW. Elements is built differently. The [RX 12 edition overview](https://docs.izotope.com/rx12/en/rx-overview.html) describes it as a set of six plug-ins with no Audio Editor, and its plug-in table lists De-ess for Standard and Advanced only. That leaves Elements owners one route: the De-Ess section inside Repair Assistant. | Where you de-ess | Editions | Controls | How you check the result | | ---------------------------------------------------- | ---------------------------- | --------------------------------------------------------------------------------------- | --------------------------------------------------------- | | De-ess module (RX Audio Editor) | Standard, Advanced | Algorithm, Threshold with Absolute, Cutoff freq, Speed, Spectral shaping, Spectral tilt | Listen with Preview, Bypass, Compare, then Render | | De-ess plug-in (DAW) | Standard, Advanced | The same documented control set | Bypass only; the other footer controls stay in the module | | Repair Assistant plug-in, Voice mode, De-Ess section | Elements, Standard, Advanced | Shaping, Threshold, Fast/Slow | The section's Audition button plays only what it removes | ![Route diagram: RX 12 Standard / Advanced leads to the De-ess module in the RX Audio Editor and the De-ess plug-in for the DAW; RX 12 Elements leads to the Repair Assistant plug-in, then Voice, then De-Ess with Shaping, Threshold and Fast/Slow](https://storage.ghost.io/c/af/4b/af4b05d4-2e0b-4edc-8378-1e2a62fecc70/content/images/2026/10/rx-de-ess-edition-route.webp) Standard and Advanced get the full De-ess in two places; Elements reaches a three-control De-Ess through Repair Assistant's Voice mode. On the plug-in side, the [RX 12 release notes](https://www.izotope.com/pages/release-notes/rx-advanced) list AU, AAX, AAX AudioSuite and VST3 formats (plus ARA for the Spectral Editor), 64-bit only, on macOS Sonoma, Sequoia and Tahoe and on Windows 10 and 11\. The Repair Assistant plug-in is narrower: AAX AudioSuite, AU and VST3\. The same notes name the supported hosts, from Logic Pro 12 and Pro Tools 2025 (Pro Tools 2026 on the current product pages) to Reaper 7 and DaVinci Resolve 19\. Sound Forge Pro is not on that list, so if you plan to stay inside Sound Forge, the manual-repair and Dynamic EQ route is in the [guide to reducing sibilance in Sound Forge](https://soundforgepro.com/reduce-sibilance-sound-forge/). RX 12.1.0, released October 15, 2026, is the current update, and neither it nor 12.0.0 lists a De-ess change. The RX 6 and RX 12 manuals describe the same De-ess controls in nearly the same words, so an older De-ess tutorial still matches the panel, though not the edition split or the host list. If you are choosing an edition mainly for voice work, De-plosive, Mouth De-click and Breath Control are also Standard and Advanced tools; the full split is in the [RX 12 edition comparison](https://soundforgepro.com/izotope-rx-elements-vs-standard-vs-advanced/). ## Classic or Spectral: pick by what you hear Spectral is the better starting point for most speech, and the manual explains why. Classic detects a sibilant and pulls the whole signal down with a broadband gain envelope, so every frequency dips for that moment. Spectral attenuates only the high frequencies where the sibilance is active and leaves the lower content untouched. iZotope describes Spectral as a multiband compressor with dozens of bands, which lets it compress the S, change its spectral shape and avoid modulating the background noise behind the voice. Classic still earns its place. iZotope's own [guide to the dos and don'ts of de-essing](https://www.izotope.com/community/blog/the-dos-and-donts-of-de-essing) calls it RX's wideband mode. When the S arrives as one hard, full-range burst, a brief dip across the whole word can sound less processed than a voice whose top end alone keeps moving. | What you hear | Start with | Why | Listen out for | | --------------------------------------------------------- | ---------------------------------------------------------------------- | ---------------------------------------------------------- | ----------------------------------------------- | | Sharp, spitty S over vowels that already sound balanced | Spectral | Only the region above Cutoff freq is turned down | The S softening into a "th" | | The whole S or SH syllable jumps out louder than its word | Classic | A short full-range dip lowers the entire event | The word ducking or losing weight | | Spectral leaves a thin whistle or a ragged top on the S | Spectral with a little more Spectral shaping, compared against Classic | Shaping changes the S's spectral shape, not only its level | Consonants flattening into dull hiss | | Classic makes the voice thin out on every S | Spectral | Lower frequencies stay where they were | A Cutoff freq set so low that vowels get caught | ![Two-panel diagram of a voice spectrum with a sibilance peak on the right: in the Classic panel the whole curve drops under a Broadband gain dip arrow; in the Spectral panel a dashed Cutoff line separates the unchanged Vowel body untouched area from the peak, which is lowered and marked Reduced above Cutoff](https://storage.ghost.io/c/af/4b/af4b05d4-2e0b-4edc-8378-1e2a62fecc70/content/images/2026/10/rx-de-ess-classic-vs-spectral.webp) Classic lowers everything while the S plays; Spectral turns down only what sits above the Cutoff and leaves the vowel body where it was. ## The RX 12 De-ess controls, one by one The labels below are the ones the RX 12 module shows. One naming quirk: the panel says Spectral shaping \[%\], while the control list on the [De-ess manual page](https://docs.izotope.com/rx12/en/de-ess.html) calls the same slider Spectral Flattening. I use the panel name here, since that is what you click. | Control | What it does | When to move it | | ---------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------- | | Algorithm (Classic, Spectral) | Chooses a broadband gain dip or frequency-specific reduction above the Cutoff | First, before anything else | | Cutoff freq \[Hz\] | The crossover between preserved speech and reduced sibilance, and the lower edge of sibilance detection | Second, while looking at the spectrogram | | Threshold \[dB\] | The level at which De-ess starts compressing sibilance | Third, lowering it until the harsh esses trigger | | Absolute (checkbox) | Unchecked is Relative, the default: the threshold follows the measured speech level. Checked sets the threshold in dB below full scale | Only for material whose level is already tightly controlled | | Speed (Fast, Slow) | Sets attack and release times; attack is program-dependent in both modes | By symptom: Fast for pumping, Slow for softened attacks | | Spectral shaping \[%\] | 0% keeps the S's natural shape with even compression across bands; 100% flattens it toward the Tilt profile and reduces it more | Only when the S that remains sounds uneven or whistly | | Spectral tilt (Brown, Pink, White) | Sets the target profile: 0 is a natural pink-noise-like slope, below 0 leans darker toward brown, above 0 brighter toward white | Last, and only with shaping above 0% | ![RX 12 De-ess module with Spectral selected under Algorithm, Threshold [dB] and Cutoff freq [Hz] faders with empty value boxes, Absolute unticked, Speed set to Fast, Spectral shaping [%] and Spectral tilt sliders with Brown, Pink and White, and footer buttons Preview, Bypass, plus, Compare, an ear icon and Render; orange markers 1 to 4 point to Spectral, Spectral shaping, Spectral tilt and the ear icon](https://storage.ghost.io/c/af/4b/af4b05d4-2e0b-4edc-8378-1e2a62fecc70/content/images/2026/10/rx-de-ess-spectral-controls.webp) Choose Spectral (1) first; touch Spectral shaping (2) and Spectral tilt (3) only after Cutoff and Threshold are set, and use the ear icon (4) to hear what you take out. Two of those sliders matter only in Spectral mode. With Classic selected, the RX 12 module shows Spectral shaping and Spectral tilt greyed out, and the manual adds that Tilt works best once shaping is above zero. That follows from how the pair works: Tilt sets the target that shaping flattens toward, so at 0% shaping it has nothing to steer. The Absolute box deserves a moment too. Relative suits most dialogue because it tracks the speech level as it rises and falls. Absolute fixes the line in dBFS. That is predictable on narration that has already been levelled, but on a recording whose level wanders, a fixed line can let quiet esses pass under it while loud passages get more reduction than they need. ## A settings order that keeps each decision audible Each step below changes one thing, so you always know which control caused what you hear. 1. Select a phrase that holds the worst esses plus the vowels around them, so you judge the S inside its words. If a handful of esses are far louder than the rest, lower those first with clip gain or a gain change on the selection. iZotope's de-essing guide gives the same advice, and it stops one extreme event from setting the threshold for the whole file. 2. Choose Spectral under Algorithm, unless the symptom table above points you to Classic. 3. Place Cutoff freq below the harsh center of the S and above the vowel body. On the spectrogram, the esses show up as bursts of high-frequency energy between the vowel bands. iZotope's guide puts sibilance typically between 4 and 10 kHz, and sometimes as low as 1.5 kHz with some voice and microphone pairings, so read the recording in front of you rather than reusing someone else's number. 4. Leave Absolute unchecked and lower Threshold until the distracting esses are caught every time. Stop as soon as words start to lisp. 5. Set Speed by ear: Fast if the high end pumps after each S, Slow if the start of each consonant loses its edge. 6. Keep Spectral shaping low. Raise it only if the S that remains sounds uneven or whistly, and move Spectral tilt after that, not before. 7. Turn on Listen, press Preview and check that you hear mostly hiss. Toggle Bypass, then press Compare to store a conservative version and a stronger one, changing a single control between them. 8. Render the version that removes the harshness while every word still sounds like itself. ## Hear what De-ess removes before you render In the module, Listen is the most useful check you have. According to the [RX 12 page on common module controls](https://docs.izotope.com/rx12/en/common-module-controls.html), Listen exists only in certain modules, and when it is on, Preview plays the signal the module would remove instead of the processed result. The RX 12 De-ess module shows the ear icon in its footer. What you want from that preview is mostly hiss. If whole syllables, vowel color or the body of a word come through, the Threshold or the Cutoff freq is too low. The rest of the footer supports the comparison. Bypass (Shift-B) flips to the untouched sound during preview. Each Compare click adds the current settings to the Compare Settings window as a new item, named after the module and numbered Settings 1, Settings 2 and so on, where you can audition them side by side and render the one you prefer. Preview also plays one second of untouched audio before and after the selection by default, and that pre- and post-roll can go up to ten seconds, which helps you hear each S inside its sentence instead of in isolation. If the RX Audio Editor itself is new to you, start with the [RX 12 beginner tutorial](https://soundforgepro.com/izotope-rx-12-tutorial-beginners/). The plug-in is more limited. The same manual page says the module footer controls are not available in the RX plug-ins, with Bypass as the only exception. In a DAW you compare by toggling Bypass at matched playback level. To hear exactly what is coming out, open the clip in the RX Audio Editor, which Standard and Advanced both include, and use the module's Listen there. In Repair Assistant, each section has its own Audition button that plays only what that section removes. ## De-essing in RX Elements with Repair Assistant If you own Elements, your de-esser is the De-Ess section of Repair Assistant. The [RX 12 Repair Assistant page](https://docs.izotope.com/rx12/en/repair-assistant.html) describes Voice mode as the mode for spoken word and dialogue, and its De-Ess section has three controls: Shaping, Threshold and Fast/Slow. No Cutoff, no Classic or Spectral choice and no Absolute option are listed. Shaping behaves like the module's slider: 0% keeps the natural shape, and 100% flattens the S toward a set spectral profile. 1. Insert the Repair Assistant plug-in on the voice track. 2. Choose Voice, click Learn and keep playing until it has built its suggestions. 3. Press Audition in the De-Ess section and listen to what it removes. 4. Adjust Threshold first, then Fast or Slow, and Shaping only if the remaining S sounds uneven. 5. Switch the section off and on to hear the difference in context. Two format details trip people up. In the AudioSuite version, the clip must be longer than seven seconds or learning appears to hang; with the AAX insert you can loop a shorter clip so it has enough to learn from. Music mode, meant for full mixes, instruments and isolated sung vocals, swaps De-Ess for De-Harsh, which has Cutoff, Threshold and Fast/Slow. Opening Repair Assistant's recommendations in the Module Chain is an Advanced-only option. If you need the Cutoff control and the choice between Classic and Spectral, you need the full De-ess in Standard or Advanced. The edition comparison shows what else that step adds, and current upgrade paths from Elements are on the [RX price, trial and license page](https://soundforgepro.com/izotope-rx-price-trial-license/). The other route is Velvet, which I cover [further down](#rx-or-velvet). ## Lisping, dull vowels, pumping: quick fixes | Symptom | Likely cause | Fix in RX | | ---------------------------------------- | --------------------------------------------------------------------- | ----------------------------------------------------------------------------------- | | The S turns into a lisp or a dull “th” | Threshold too low, or Spectral shaping too high | Raise Threshold, lower Spectral shaping, then compare a slightly higher Cutoff freq | | Vowels go dull and the voice gets darker | Classic is dipping the whole word, or Cutoff freq sits too low | Switch to Spectral and raise Cutoff freq | | The high end pumps after each S | Recovery is too slow, so high frequencies stay down too long | Choose Fast | | Consonant attacks lose their edge | Settings are fast enough to cut the start of each transient | Choose Slow | | Some esses slip through | They sit below the Cutoff, or the level changes from phrase to phrase | Check where those esses sit before lowering Threshold, and keep Relative mode | | A few extreme esses survive any setting | One pass is being asked to do too much | Lower those events with clip gain or automation, then run a lighter De-ess pass | The Fast and Slow rows come straight from iZotope's own explanation of Speed: too slow and the high end recovers late, too fast and the initial transient gets cut. If the De-ess plug-in does not show up in your DAW at all, that is an installation or scan problem rather than a settings one; the checklist for [RX plug-ins not showing](https://soundforgepro.com/izotope-rx-plugins-not-showing/) covers it. ## Where De-ess sits in a voice cleanup chain Sibilance is one of several mouth-and-microphone problems, and each has its own fix. A low thump on a P or B is a plosive and needs a [different repair for plosives](https://soundforgepro.com/remove-plosives-from-audio/); short ticks between words are [mouth clicks](https://soundforgepro.com/remove-mouth-clicks-from-audio/); noise or room sound around the voice is a job for [Dialogue Isolate](https://soundforgepro.com/izotope-rx-dialogue-isolate/). In RX Standard and Advanced, De-plosive, [Mouth De-click](https://soundforgepro.com/izotope-rx-mouth-de-click/) and Breath Control are separate modules, so give each event its own tool instead of pushing De-ess harder. On order, iZotope's de-essing guide is relaxed: it treats EQ, compression and de-essing as interdependent and accepts a de-esser before the EQ, after the compressor, first or last. Its firmer advice is to automate the de-esser and use several light stages instead of one heavy one. It also notes that a singer may sibilate higher when singing softly and push the esses harder around 3 kHz when belting, which is a good reason to treat those passages separately. For spoken word, my recommended order is to fix events such as plosives and clicks first, then de-ess, then shape tone and level. Check the esses again after compression and before loudness normalization, because later gain changes can bring leftover sibilance forward. If you deliver to a [podcast loudness standard](https://soundforgepro.com/podcast-loudness-standard/), make that check before the final limiter, and the [voice-over recording workflow](https://soundforgepro.com/sound-forge-pro-for-voice-over-recording/) shows where the rest of the processing fits. A finished stereo mix is a different job, because a de-esser on the whole mix also catches hi-hats and cymbals. In a 2020 iZotope post on [de-essing a master](https://www.izotope.com/community/blog/de-essing-a-master-tools-tips-and-tricks), Oscar Zambrano lists three options: highlight the problem region in RX and lower it with Gain or [Spectral Repair](https://soundforgepro.com/izotope-rx-spectral-repair/), use Ozone's mid/side Dynamic EQ for side-panned percussion, or export a vocal with [Music Rebalance](https://soundforgepro.com/izotope-rx-music-rebalance/) and use it as the sidechain key for a de-esser. ## Test results: what reviewers report about RX de-essing No outlet has published numeric sibilance-reduction measurements for RX De-ess, so this composite is a counted summary of dated verdicts from three independent reviews, with iZotope's own guidance kept apart. The picture is consistent: all three are positive, and the two RX reviews agree on speed and targeting. - **2 of 3 reviews say the tools cut down manual work.** [Sound On Sound's RX6 review](https://www.soundonsound.com/reviews/izotope-rx6) (July 2017, when De-ess arrived in RX6 Standard) found that the new vocal modules, De-Ess included, make jobs that once needed time and a lot of skill "a lot quicker and much easier by being expertly targeted". [Production Expert's RX 11 review](https://www.production-expert.com/production-expert-1/izotope-rx-11-review-deep-dive-from-post-expert) (May 2024) needed fewer tweaks to Repair Assistant's suggestions, with results closer to manual settings; that remark covers Repair Assistant as a whole, not its De-Ess section alone. - **2 of 3 call the tools very effective.** Sound On Sound said RX6's new modules "target specific audio problems very effectively indeed", and its [Velvet review](https://www.soundonsound.com/reviews/izotope-velvet) (July 2025) found Velvet "very effective on a range of spoken and sung vocal recordings". - **1 of 3 covers the Elements route directly.** Production Expert reported that RX 11 improved Repair Assistant's De-Ess section with additional controls; it now has Shaping, Threshold and Fast/Slow, the same three controls the RX 12 manual lists. - **The only published number belongs to Velvet.** Sound On Sound saw around 100 ms of latency in its Cubase system, which says nothing about RX De-ess. - **Vendor guidance, counted separately.** iZotope's 2025 de-essing guide says RX "has a spectral de-esser that sounds fantastic" and recommends light, staged reduction. | Review | Date | Version | Verdict on de-essing | | -------------------------------- | ------------- | ------------------------------ | ----------------------------------------------------------------------------------------- | | Sound On Sound (Hugh Robjohns) | July 20, 2017 | RX6 Standard and Advanced | Expertly targeted; quicker and easier than manual work | | Production Expert (Paul Maunder) | May 15, 2024 | RX 11, Repair Assistant | De-Ess improved with added controls (Shaping, Threshold, Fast/Slow); fewer tweaks overall | | Sound On Sound (John Walden) | July 24, 2025 | Velvet 1.0, a separate product | Very effective on spoken and sung vocals; about 100 ms latency in Cubase | The 2017 verdict still applies to the module's design, because the RX 12.0.0 and 12.1.0 release notes list no De-ess change and the RX 6 and RX 12 manuals describe the same controls. My verdict: for recorded speech in RX Standard or Advanced, Spectral De-ess with a Relative threshold is the right default, and Listen plus Compare make it easy to stop before the voice starts to lisp. Elements owners get a usable, simplified version in Repair Assistant, without Cutoff or a mode choice. If you want a real-time de-esser on a vocal insert while you mix, Velvet is the alternative. For how De-ess fits the rest of the package, see the [RX 12 review](https://soundforgepro.com/izotope-rx-12-review/). ## RX De-ess or Velvet? Velvet is a separate iZotope plug-in in the Catalyst Series, released as version 1.0.0 on May 6, 2025 according to its [release notes](https://www.izotope.com/pages/release-notes/velvet). It is not part of any RX edition, and it approaches the job differently. | | RX De-ess | Velvet | | ------------------ | --------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------- | | Best fit | Repairing recorded dialogue, then rendering the fix into the file | A vocal insert at the mix stage | | Where it runs | RX Audio Editor module and DAW plug-in | DAW plug-in: AAX, AU and VST3, 64-bit only | | How it finds the S | Threshold and Cutoff freq, in Classic or Spectral mode | Sibilance Learn sets up a 6-band dynamic EQ | | Extras | Listen, Compare and Render in the module | Tonal channel with Lift and Tame across 8 target profiles, De-Click, 50 presets, a Delta button and a Mix fader | | Latency note | No published figure for the plug-in; the module renders into the file | About 100 ms in Sound On Sound's Cubase system | | How you get it | RX 12 Standard or Advanced | Separate purchase: $49 US list price in October 2026 (EDU $24.50) | I'd reach for RX De-ess for dialogue repair, where hearing the removed signal and rendering a fixed file matter most. Velvet makes more sense as a mix-stage insert, including for Elements owners who want a full de-esser in the DAW. Sound On Sound judged its latency fine for mixing but best avoided while tracking. Feature details are on the [Velvet product page](https://www.izotope.com/products/velvet), RX pricing stays on the price page, and for tools outside iZotope, see the roundup of [iZotope RX alternatives](https://soundforgepro.com/best-izotope-rx-alternatives/).