> ## 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. # How to Clean Vocals in iZotope RX 12: The Repair Order, Step by Step - URL: https://soundforgepro.com/clean-vocals-izotope-rx/ - Published: 2026-10-27T17:47:00.000Z - Updated: 2026-10-27T17:46:59.000Z - Description: A six-step RX 12 vocal cleanup order: clipping and clicks first, mouth noise and plosives next, then hum, noise and room, with breaths and sibilance last. Which edition each module needs, which placements are fixed rules, and what the published reviews agree on. - Author: Erick Finn - Tags: iZotope RX, Voice Recording **To clean vocals in iZotope RX 12, repair the deepest damage first and the most cosmetic problems last. Fix clipping with De-clip. Remove clicks with De-click, then crackle with De-crackle. Treat mouth noise with Mouth De-click and pops with De-plosive, before any high-pass filter. Take out hum with De-hum, then steady noise with Voice De-noise set to Music or a learned Spectral De-noise profile. Reduce the room with De-reverb only if it distracts. Finish with Breath Control and De-ess, and try them in both orders. Work on a copy and judge every step by listening to what the module removes. RX 12 Elements covers clipping, clicks, hum, steady noise and room; Mouth De-click, De-plosive, De-crackle, Spectral De-noise, Breath Control and De-ess need Standard or Advanced.** My rule is that only a few steps have a fixed position. The documentation pins down three things: De-plosive goes before a high-pass filter, De-crackle works better after De-click, and the modules that learn or detect need enough material to work with. Everything else is convention, and iZotope's own writers order some steps differently. So I give my default order below and say where I would flip it. The steps follow RX 12.1.0, which the [RX release notes](https://www.izotope.com/pages/release-notes/rx-advanced) list as the newest version, released on October 15, 2026. This page assumes the vocal is already on its own track: a home-studio take, a comp or a live vocal mic. If the singer is still inside a finished song, separate the voice first. [Stem Split opens the separated stems in a dedicated Stems tab](https://docs.izotope.com/rx12/en/stem-split-pane.html), and the guide to [Music Rebalance and Stem Split](https://soundforgepro.com/izotope-rx-music-rebalance/) covers that job. Then come back to this order. Before you open a module, play the vocal in a rough mix and write down the problems you can hear, then listen to it soloed and add to the list. iZotope's vocal guide starts with these two passes so the cleanup doesn't chase problems the mix would hide. I would skip any row of the table below that isn't on your list. ## The RX 12 vocal cleanup order at a glance Read the table from top to bottom. Each row is one step, the sound that tells you it is needed, the module that fixes it and the cheapest edition that has that module. The last column separates the documented rules from the habits. | Step | What you hear | RX 12 module | Lowest edition | Rule or convention | | ---------------------------- | -------------------------------------------------------------- | -------------------------------------------------------- | --------------------------------------------------------------- | --------------------------------------------------------------------------------------- | | 1\. Clipping | Flat-topped, crunchy peaks on the loudest notes | De-clip | Elements | Convention: the deepest damage goes first | | 2\. Clicks and crackle | Sharp ticks, or a fine, dense crackle under the voice | De-click, then De-crackle | Elements (De-click), Standard (De-crackle) | Rule: De-crackle after De-click | | 3\. Mouth noise and plosives | Wet clicks and lip smacks between words; low thumps on P and B | Mouth De-click, then De-plosive | Standard | Rule: De-plosive before any high-pass filter. Its place in the sequence is a convention | | 4\. Hum and steady noise | A low tone or buzz, then hiss, fan or air-conditioning noise | De-hum, then Voice De-noise (Music) or Spectral De-noise | Elements (De-hum, Voice De-noise), Standard (Spectral De-noise) | Convention: tonal noise before broadband noise | | 5\. Room | Ringing tails, a boxy or distant sound | De-reverb | Elements | Only when the room distracts | | 6\. Breaths and sibilance | Loud inhales, sharp S and T sounds | Breath Control, De-ess | Standard | Convention: try both orders | The logic behind the sequence comes from [Mike Metlay's article on repair order](https://www.izotope.com/community/blog/order-of-audio-repair-operations): tackle the deepest damage first, starting with clipping, and treat variable noise such as plosives, sibilance and unwanted reverberance as the shallowest layer. [iZotope's own vocal cleanup guide](https://www.izotope.com/community/blog/clean-up-vocals) runs a similar but not identical sequence. It starts with Mouth De-click and De-plosive, handles clicks, crackle, clipping and Breath Control next, and leaves hum, reverb and broadband noise for later. The [iZotope RX hub](https://soundforgepro.com/izotope-rx/) groups the module guides by problem if you need more depth on one step. ![Six-step RX 12 vocal cleanup order from clipping to breaths and sibilance, with the modules for each step colored by the lowest edition that includes them](https://storage.ghost.io/c/af/4b/af4b05d4-2e0b-4edc-8378-1e2a62fecc70/content/images/2026/10/vocal-cleanup-order.webp) Deepest damage first, cosmetic fixes last. Only two of the placements are fixed rules. ## What Elements, Standard and Advanced can do on a vocal Check your edition before you plan the chain, because two of the six steps depend entirely on modules that Elements does not have, and two more lean on one. [iZotope's RX 12 module and plug-in tables](https://docs.izotope.com/rx12/en/rx-overview.html) split the work like this: - **Elements:** six DAW plug-ins and no standalone Audio Editor. The six are De-click, De-clip, De-hum, De-reverb, Repair Assistant and Voice De-noise. Voice De-noise is a plug-in only in this edition. - **Standard:** adds the Audio Editor and the vocal modules: Mouth De-click, De-plosive, De-crackle, Spectral De-noise, Breath Control and De-ess, each as a module and a plug-in. It also adds Spectral Repair and Interpolate, which exist only inside the Audio Editor. - **Advanced:** adds Deconstruct and EQ Match, a high-quality offline mode and multiband processing in Dialogue Isolate, and the Open Module Chain button in Repair Assistant. An Elements owner can still do a real first pass: De-clip, De-click (its Low latency algorithm handles mouth clicks), De-hum, Voice De-noise, De-reverb and Repair Assistant. What Elements cannot do is target plosives, breaths or sibilance with a dedicated module. Reviewers agree on that split, as the test results further down show. For everything beyond vocals, the [RX 12 Elements vs Standard vs Advanced comparison](https://soundforgepro.com/izotope-rx-elements-vs-standard-vs-advanced/) lists every module, and current prices, the trial and upgrade paths are in the [RX price and licensing guide](https://soundforgepro.com/izotope-rx-price-trial-license/). ## Set up so every step can be undone Three decisions come before the first module: protect the original, pick where the processing happens and set up your monitoring. **Protect the original.** The [RX file-handling documentation](https://docs.izotope.com/rx12/en/working-with-files.html) warns that saving an uncompressed file (.wav or .aiff) in the Audio Editor overwrites the original on disk. Duplicate the take, or the playlist in your DAW, before you open it in RX. The raw vocal should sit somewhere RX never writes to. **Pick the place.** RX 12 plug-ins are 64-bit only. In a DAW you do not have to insert RX across the whole track. [iZotope's article on RX in music production](https://www.izotope.com/community/blog/rx-music-production) suggests processing only the problem region instead: AudioSuite in Pro Tools, Selection Based Processing in Logic or item effects in Reaper. Spectral Repair and Interpolate are the exception, because they live only in the Audio Editor. In Standard and Advanced, the Spectral Editor plug-in also works over ARA in Logic Pro 10.7 or later (AU, Rosetta only on Apple silicon) and in Studio One 7 or Studio Pro 8 (VST3). If the plug-ins are missing from your DAW, see [RX plug-ins not showing up](https://soundforgepro.com/izotope-rx-plugins-not-showing/). **Monitor the removed signal.** With Listen active, [Preview plays the signal a module would remove](https://docs.izotope.com/rx12/en/common-module-controls.html) instead of the processed result. In RX 12, most processing modules also offer difference monitoring through the ear button. My rule: if you can hear words, pitch or the outline of a note in the removed signal, the module is taking voice with the noise, so back off. If selections, Preview and Render are new to you, start with the [RX 12 beginner tutorial](https://soundforgepro.com/izotope-rx-12-tutorial-beginners/). ## The fast route: Repair Assistant in Music mode Repair Assistant is in every edition, including Elements, and it is the quickest way to get a first suggestion for a whole vocal. Open it, choose a mode and click Learn. The [Repair Assistant documentation](https://docs.izotope.com/rx12/en/repair-assistant.html) sets out how it behaves: - **Choose Music mode for singing.** Music mode is intended for full mixes, individual instruments or isolated sung vocals. Voice mode is intended for spoken word or dialogue, and the choice changes the processing chains it generates. - **No noise-only section needed.** Unlike Spectral De-noise, Repair Assistant does not need to learn on a stretch of pure noise. - **Mind the clip length in AudioSuite.** As an AudioSuite plug-in it needs clips longer than 7 seconds, or it appears to learn indefinitely. In AAX real-time use you can loop a shorter clip. - **Audition each section.** Every section has an Audition button that plays only what that section removes. - **Read what it switched on.** Its documented processing sections include De-ess and De-Harsh alongside the noise and click tools, so check which ones it used on your take. In Advanced, the Open Module Chain button sends the suggestion to [Module Chain](https://docs.izotope.com/rx12/en/module-chain.html), which runs several modules in series and can save the chain as a preset. In Standard and Elements you adjust the suggested settings inside the assistant instead. I treat the result as a draft, not a finished vocal. Compare each section it enabled with the matching step below. Ask whether it is doing that step's job, and whether its removed signal contains any voice. For an Elements owner, or for a quick demo, the draft may be all the cleanup the take needs. ## Steps 1 and 2: clipping, clicks and crackle I'd start with clipping because it is the deepest damage on the take. Every later module has to work on whatever De-clip leaves behind. The [De-clip documentation](https://docs.izotope.com/rx12/en/de-clip.html) gives three practical points: - Set Threshold just below the actual level of the clipping. - Preview the Low quality modes first and move up only if they fall short. - If you disable the post-limiter, restored peaks above 0 dBFS are kept without makeup gain only when you save the file as 32-bit float. Clicks come next. In [RX 12 De-click](https://docs.izotope.com/rx12/en/de-click.html), the Multi-band (Random clicks) algorithm includes a protective algorithm that preserves periodic elements of instruments such as brass or vocals, which makes it my first choice on a sung take. The Low latency algorithm works well on mouth clicks and on clicks the other algorithms cannot handle, which matters for Elements owners in the next step. If a fine crackle is left after De-click, reach for De-crackle next. iZotope's own De-crackle documentation says the module is even more effective once De-click has removed the worst clicks. That is one of the few fixed placements in the whole order. De-crackle needs Standard or Advanced. ## Step 3: mouth noise and plosives [Mouth De-click](https://docs.izotope.com/rx12/en/mouth-de-click.html) is the dedicated tool for wet clicks and lip smacks, available as a module and plug-in in Standard and Advanced only. Three documented points shape how I'd set it: - **Sensitivity cuts both ways.** Raising it can affect plosives and reduce or damage the original signal. - **Click Widening covers smacks.** It extends the repair area around each detected click, which takes care of lip smacks that have a decay. - **Two passes can beat one.** Running Mouth De-click twice sometimes works better, because the first pass can miss quieter clicks that louder ones masked. So keep Sensitivity moderate and run a second light pass rather than one hard pass. On Elements, use De-click's Low latency algorithm for mouth clicks instead. [De-plosive](https://docs.izotope.com/rx12/en/de-plosive.html) detects plosives between 20 and 80 Hz. If the audio has already gone through a high-pass filter, that detection may not work correctly, so iZotope recommends running De-plosive before any high-pass filter. That is the one hard rule in this step. Set the three controls in this order: 1. **Sensitivity** decides how much of the input counts as a plosive. It has a bigger effect on the result than Strength. 2. **Strength** sets how much the detected plosives are reduced. High values cut more but cost clarity. To keep the voice while still reducing pops, raise Sensitivity and lower Strength. 3. **Frequency limit \[Hz\]** is the upper boundary of the reduction. Plosives typically sit between 20 and 300 Hz and can reach 500 Hz, so set the limit just above where the pops in your take occur. ![RX 12 De-plosive panel with Sensitivity, Strength and Frequency limit numbered 1 to 3, above a frequency ruler marking the 20 to 80 Hz detection band and the typical 20 to 300 Hz plosive range extending toward 500 Hz](https://storage.ghost.io/c/af/4b/af4b05d4-2e0b-4edc-8378-1e2a62fecc70/content/images/2026/10/de-plosive-controls.webp) Raise Sensitivity before Strength, and keep the limit just above the pops. Where De-plosive sits in the sequence is a convention. iZotope's vocal guide fixes plosives early, right after mouth clicks. Metlay treats plosives as variable noise and repairs them last. I take the early slot, next to the mouth noise, because De-plosive needs to see untouched low end. Without Standard, [removing plosives in Sound Forge](https://soundforgepro.com/remove-plosives-from-audio/) covers the manual route with a filter and a selection. ## Step 4: hum, hiss and steady noise Hum is a tone, so I remove it before the broadband noise. [De-hum](https://docs.izotope.com/rx12/en/de-hum.html) has a Static mode for ground noise or a simple hum with a few harmonics and a Dynamic mode for complex hum or buzz. In the plug-in, its Adaptive Dynamic mode is not optimized for real-time use, so render it rather than monitoring through it. iZotope's guide to [identifying audio problems](https://docs.izotope.com/rx12/en/identifying-audio-problems.html) describes hum as a low tone based at 50 or 60 Hz, depending on where the recording was made. When the harmonics extend above 400 Hz, the same page says Spectral De-noise is often more effective than De-hum. For steady hiss, fan or air conditioning, [Voice De-noise](https://docs.izotope.com/rx12/en/voice-de-noise.html) is the zero-latency option in every edition. On a sung vocal, set it up like this: - **Optimize for Music.** It does not attenuate sustained notes and is more transparent on singing. Dialogue reacts faster to changing noise but is not meant for the noise in sung vocals. - **Learn from at least one second of pure noise** if you use Manual mode. - **Start with Gentle.** Surgical removes more noise but sacrifices some timbre and can leave musical-noise artifacts. When the noise is heavier, or has a clear shape you want to target, Spectral De-noise with a learned profile gives more control. It needs Standard, and its plug-in is far more resource intensive than Voice De-noise, with higher latency. The [Spectral De-noise documentation](https://docs.izotope.com/rx12/en/spectral-de-noise.html) says to learn from the longest noise-only section available, ideally a few seconds long. Watch Artifact control: under heavy processing, low values can sound chirpy or watery, and high values can sound like broadband gating. iZotope also describes Spectral De-noise as usable on singing voice, not only on dialogue. Two more tools fit here. Advanced owners have [Deconstruct](https://docs.izotope.com/rx12/en/deconstruct.html), which splits a selection into tonal, noisy and optionally transient components, each with its own gain. If the noise is headphone or monitor spill rather than hiss, [De-bleed offers a Vocals signal type](https://docs.izotope.com/rx12/en/de-bleed.html). Module-by-module noise settings, artifacts and listening results are in the guide to [removing background noise with iZotope RX](https://soundforgepro.com/remove-background-noise-izotope-rx/). ## Step 5: room sound, only when it distracts A dry vocal is not automatically a better one. A little room often sits fine once the track is in the mix, so I reduce the room only when it pulls attention away from the singer. For a sung vocal, [De-reverb](https://docs.izotope.com/rx12/en/de-reverb.html) is the module, and it is in every edition. Its documentation gives the setup: - **Learn on the right selection.** Use several seconds that start with some noise floor or room tone and include both the direct voice and its reverb tails. - **Read Tail length as decay time.** It approximates RT-60, the time a reverberant signal takes to drop by 60 dB. - **Turn on Output reverb only** to hear just the reverb you are taking away. If the voice itself shows up there, back off. - **Use the high band on live vocals.** Raising the high-band control can curtail ringing sibilance in live vocal recordings. - **Pair it on very reverberant takes.** iZotope says De-reverb combined with Spectral De-noise can tame very reverberant signals. Speech is a different case. [Dialogue Isolate](https://docs.izotope.com/rx12/en/dialogue-isolate.html) separates dialogue, noise and reverberation into three components with independent Voice, Reverb and Noise gains, and the Advanced version adds a Best/Offline mode with fewer artifacts at the cost of longer processing. The same page points general, non-dialogue reverb reduction to De-reverb, which is why De-reverb leads here. For podcasts and voice-over rather than singing, the [Dialogue Isolate guide](https://soundforgepro.com/izotope-rx-dialogue-isolate/) covers the gains. ## Step 6: breaths, sibilance and spot fixes Breaths and sibilance come last because they are the most cosmetic, and how much to keep is partly taste. [Breath Control](https://docs.izotope.com/rx12/en/breath-control.html) uses machine learning to detect breaths in dialogue or vocal recordings regardless of their level. With the Real-time or Offline algorithm, detection works better on longer selections, preferably at least 3 seconds. Then make four decisions: 1. **Algorithm.** Real-time is the fastest and suits live plug-in use. Offline gives the best results but uses more resources, so save it for the final pass. Classic is the algorithm from RX 11\. What else changed since then is in [RX 12 vs RX 11](https://soundforgepro.com/izotope-rx-12-vs-rx-11/). 2. **Target or Gain.** Target mode brings each detected breath down only to the Level setting, so quiet breaths stay untouched and the result sounds more natural. Gain mode cuts every breath by the same amount, which can sound unnatural. Target is my default. 3. **Natural or Gated.** Natural preserves the ambience around the breath; Gated removes it. The switch is unavailable with the Classic algorithm. 4. **Level.** It runs from -Inf (full removal) to 0 dB (no reduction), so Breath Control never boosts a breath. Lower breaths rather than delete them, because the phrase still has to sound sung. Use the ear button while you set these. If you can hear the start of a word in the removed signal, the detection is too aggressive for that passage. ![RX 12 Breath Control panel with the Algorithm menu, Target and Gain switch, Natural and Gated switch, Level fader and ear button numbered 1 to 5](https://storage.ghost.io/c/af/4b/af4b05d4-2e0b-4edc-8378-1e2a62fecc70/content/images/2026/10/breath-control-controls.webp) Target mode trims only the loud breaths; Natural keeps the room around them. [De-ess](https://docs.izotope.com/rx12/en/de-ess.html) has two modes. Classic mode attenuates sibilants with a broadband gain envelope. Spectral mode attenuates only the high frequencies where sibilance is active, and is technically a multiband compressor with dozens of bands. Cutoff sets the lower boundary for sibilance detection, so it decides where the voice ends and the S begins. Which goes first is a convention. In Metlay's dialogue example, Breath Control before De-ess produced artifacts at phrase ends, and swapping to De-ess first worked better, so try both orders on a copy. Note that iZotope's vocal guide leaves de-essing out entirely. I'd de-ess in RX only when the sibilance is a recording defect that every later process will exaggerate. Otherwise leave de-essing to the mix, where it can be set against the final EQ. Some defects survive every pass: a single smack, a chair creak inside a note, one stray click. Two Audio Editor modules handle these, both in Standard and Advanced only. [Spectral Repair](https://docs.izotope.com/rx12/en/spectral-repair.html)'s Attenuate mode pulls a selection down to match its surroundings without resynthesizing audio. [Interpolate](https://docs.izotope.com/rx12/en/interpolate.html) repairs individual clicks shorter than 4000 samples. Without Standard, [editing breaths in Sound Forge](https://soundforgepro.com/edit-breaths-sound-forge/) does the breath work by hand. ## What the RX 12 reviews say about the vocal tools This composite result draws on four independent RX 12 reviews. None of them published vocal-specific measurements such as decibels of breath reduction or artifact scores, so the composite counts what each reviewer reports about the tools a vocal cleanup relies on. All four reviews treat Standard as the realistic tier for a full cleanup toolkit, and three of them describe Elements as a small set of six plug-ins. Three of four single out Breath Control as rebuilt or improved. Two of four say the new ear (difference) monitoring makes it easier to hear what a module removes. Two of four also report that the rebuilt tools work faster without a cost in quality, and three of four note an improved Dialogue Isolate for the speech side. | Review | Date | What it reports for vocal cleanup | | ---------------------------------------------------------------------------------------------------------------------- | --------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | [MusicTech's RX 12 review](https://musictech.com/reviews/plug-ins/izotope-rx-12-review/) (Adam Crute) | April 29, 2026 | The machine-learning Breath Control gives faster setup and more accurate breath recognition and removal than before. Most processing modules now offer difference monitoring. | | [Sound On Sound's RX 12 Advanced review](https://www.soundonsound.com/reviews/izotope-rx12-advanced) (Hugh Robjohns) | July 2026 issue | The ear button's delta monitoring makes it far easier to judge when processing is optimised. Many rebuilt tools seem to work faster and more precisely without sacrificing audio quality. Elements is six relatively basic but capable noise-removal tools; Standard has over 70. | | [Creative Field Recording's look at RX 12](https://www.creativefieldrecording.com/2026/05/06/rx-12-for-sound-effects/) | May 6, 2026 | Lists Breath Control among the modules improved in RX 12. | | [MixingGPT's RX 12 review](https://mixinggpt.com/blog/izotope-rx-12-review-2026) | June 19, 2026 | Says Breath Control was rebuilt with new neural nets for finer accuracy and faster setup, and calls a moderate reduction the sweet spot. Its suggestion that Breath Control can boost breaths contradicts the RX 12 documentation, where Level stops at 0 dB. | My verdict: RX 12 Standard is the edition that actually covers a vocal cleanup, because the steps that separate a clean vocal from a merely quiet one (mouth noise, plosives, breaths and sibilance) all need it. The rebuilt Breath Control and the ear monitor are the two RX 12 changes that matter most for singers: one does the most tedious edit, and the other shows you when a module is taking voice with the noise. Elements still handles a solid basic pass on clipping, clicks, hum, steady noise and room, with Repair Assistant as a starting point. The full verdict is in the [iZotope RX 12 review](https://soundforgepro.com/izotope-rx-12-review/). ## Check the cleaned vocal in the mix, then hand it off Solo listening is for finding problems. The mix is where you judge the fix. iZotope's vocal guide also recommends keeping the original, untouched file when you export, so the comparison is always available. Before the vocal leaves RX, run through these checks: - Play the cleaned vocal against the untouched original, inside the song, at mix level. - Listen to the removed signal once more for each module. Words, pitch or consonant edges in it mean that step went too far. - Check the P, B, S and T sounds. A soft, dull consonant or a slight lisp means De-plosive or De-ess is too strong. - Check the phrase ends. Clipped tails or pumping ambience after Breath Control point to the module order or to Gated mode. - Check the sustained notes. A watery or chirpy tail means the noise reduction went too deep. - Check the room. If the voice sounds thin or hollow after De-reverb, ease it off. Then save in two forms. An RX Document (.rxdoc) keeps your work for later revisits, but it opens only in the RX Audio Editor. For the session, bounce a WAV under a new name and never over the original. Compression, EQ, tuning and creative de-essing belong to the mix, not to RX. Once the take is free of defects, the repair job is done.