> ## 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 Remove Background Noise in iZotope RX 12: Modules and Settings - URL: https://soundforgepro.com/remove-background-noise-izotope-rx/ - Published: 2026-10-21T07:41:00.000Z - Updated: 2026-10-22T01:08:57.000Z - Description: Pick the RX 12 module by noise type, then set it in order: Voice De-noise for a steady floor, Spectral De-noise with a learned profile, Dialogue Isolate for noise that keeps changing, De-hum and Spectral Repair for hum and one-off sounds. Edition limits, prices and what published shootouts found. - Author: Erick Finn - Tags: iZotope RX, Sound Forge Restoration Guides **To remove background noise in iZotope RX, match the module to the noise. Use Voice De-noise for steady hiss, fan or air-conditioning noise in any edition, or Spectral De-noise with a learned profile when you want finer control (Standard and Advanced). Use Dialogue Isolate for traffic, crowds and other changing noise behind speech (Standard and Advanced), De-hum for hum and buzz, and Spectral Repair for one-off sounds such as a door slam (Standard and Advanced, editor only). Before you commit, listen to what is being removed, and keep the settings light: a couple of gentle passes sound more natural than one heavy pass.** My rule is to start with the lightest tool that fits the noise. Voice De-noise runs with zero latency and needs almost no setup, so it is where I would start for a steady floor under a voice. I would move up to Spectral De-noise when the noise has a clear, learnable profile and I want separate control over its hiss and its tonal parts. I would reach for Dialogue Isolate only when the noise changes from moment to moment, because that is the job its machine-learning model was built for. Below are the settings for each module, the artifacts that tell you to back off, how RX fared in published listening tests and which edition you need. ## Which RX module removes which background noise [iZotope's guide to cleaning up audio](https://www.izotope.com/community/blog/how-to-clean-up-audio-and-remove-background-noise) splits noise into two families. Steady noise such as hiss, hum and buzz suits Spectral De-noise or Voice De-noise. Sudden, intermittent noise such as a phone ring or a door squeak is better handled with Spectral Repair. Dialogue Isolate adds a third route for speech buried in noise that keeps changing. The table maps the common cases to a first module and the cheapest edition that includes it. | Noise you hear | First module | Minimum edition | Plug-in or editor | Why it fits | | ---------------------------------------------------------------------------------------------- | --------------------------- | --------------- | ---------------------------------------------------------------- | ------------------------------------------------------------- | | Steady hiss, fan, air conditioning | Voice De-noise | Elements | Plug-in in Elements; module and plug-in in Standard and Advanced | Zero-latency multiband gate with an Adaptive mode | | The same steady noise when you want finer control (tape hiss, HVAC, camera motors, line noise) | Spectral De-noise | Standard | Module and plug-in | Learns a profile of stationary tonal noise and broadband hiss | | Traffic, crowds, footsteps or weather behind speech | Dialogue Isolate | Standard | Module and plug-in | Separates dialogue from constant or changing noise | | Mains hum and buzz | De-hum | Elements | Plug-in in Elements; module and plug-in in Standard and Advanced | Static mode for simple hum, Dynamic mode for complex buzz | | A door slam, chair squeak or similar one-off sound | Spectral Repair (Attenuate) | Standard | Editor module only | Pulls the event down to the level of its surroundings | | Light to moderate wind bursts on the mic | De-wind | Advanced | Editor module | Built for wind bursts, not for wind that distorts the mic | The two routes overlap on stationary noise. For hiss, buzz or line noise, the [Dialogue Isolate documentation](https://docs.izotope.com/rx12/en/dialogue-isolate.html) suggests trying Spectral De-noise if Dialogue Isolate does not give acceptable results. New to selections, Preview and the editor layout? Start with the [RX 12 beginner tutorial](https://soundforgepro.com/izotope-rx-12-tutorial-beginners/), because the steps below assume you can make a selection and audition a module. ![Decision chart matching steady hiss, changing noise behind speech, hum, one-off events and wind bursts to the RX 12 module and minimum edition for each](https://storage.ghost.io/c/af/4b/af4b05d4-2e0b-4edc-8378-1e2a62fecc70/content/images/2026/10/noise-type-decision.webp) Name the kind of noise first, then check that your edition has the module. ## RX Voice De-noise: the fast route for steady noise Voice De-noise is the noise reducer that every RX 12 edition includes. According to the [RX 12 Voice De-noise documentation](https://docs.izotope.com/rx12/en/voice-de-noise.html), it exists only as a plug-in in Elements and as both a module and a plug-in in Standard and Advanced. It works as 64 psychoacoustically spaced band-pass filters acting as a multiband gate. Anything above the threshold in a band passes untouched, and anything below it is turned down. That design is why it runs with zero latency, while the Spectral De-noise plug-in is far more resource-intensive and adds latency. 1. **Insert it.** Put the plug-in on the dialogue or vocal track in your DAW or NLE, or open the module in the RX editor if you have Standard or Advanced. 2. **Pick Optimize for Dialogue or Music.** Dialogue reacts to noise changes faster and is not meant for the noise in sung vocals. Music does not attenuate sustained notes and is more transparent on singing. 3. **Choose Adaptive or Learn.** Adaptive mode follows a changing noise floor and keeps its threshold lower to prevent artifacts. For a fixed floor, use Manual mode and Learn from at least one second of pure noise. 4. **Start with the Gentle filter type.** Gentle is more transparent but leaves more high-end sizzle. Surgical removes more noise, at some cost to timbre, and can produce musical-noise artifacts. 5. **Set Threshold, then Reduction.** The master Threshold moves all six threshold nodes together; Reduction sets the maximum depth of reduction per band in dB. Raise Reduction until the noise stops drawing attention, not until it vanishes. ![RX 12 Voice De-noise window with Adaptive mode, Optimize for Dialogue, Gentle filter type and the Threshold and Reduction controls numbered 1 to 4](https://storage.ghost.io/c/af/4b/af4b05d4-2e0b-4edc-8378-1e2a62fecc70/content/images/2026/10/voice-de-noise-controls.webp) Pick the mode and filter type before touching Reduction. Set your expectations to match the design. [ProVideo Coalition's RX 11 shootout](https://www.provideocoalition.com/izotope-rx-11-shootout/) left Voice De-noise out of its dialogue comparison because it "doesn't really compete" with the machine-learning tools, while calling it great for minor ambience reduction. That is the right job for it: a light, steady floor, not a busy street. ## RX Spectral De-noise settings that keep the voice natural Spectral De-noise learns a profile of stationary or slowly changing noise, such as tape hiss, HVAC, fans, camera motors or line noise, and subtracts it when the signal drops below the threshold. The [RX 12 Spectral De-noise documentation](https://docs.izotope.com/rx12/en/spectral-de-noise.html) also explains that it splits noise into tonal parts (hum, buzz, interference) and random parts (hiss), with separate Threshold and Reduction settings for each. Work through the panel in this order. 1. **Learn from the longest clean stretch.** Select the longest noise-only section you can find, ideally a few seconds. These usually sit at the start or end of the file or in pauses between words. Click Learn in the editor module; in the plug-in, engage Learn and play the noise back, or use Preview in AudioSuite. If no single stretch is long enough, the standalone application can build the profile from several isolated selections. 2. **Start Threshold at 0 dB.** A higher threshold removes more noise but also suppresses low-level detail. A lower one keeps detail but can let the noise be modulated by the signal. If the noise rises and falls over time, like traffic or record surface noise, iZotope suggests raising Threshold. 3. **Raise Reduction only as far as you need.** Strong suppression degrades low-level signal as well, so stop where the noise becomes unobjectionable. [MixingGPT's RX 12 review](https://mixinggpt.com/blog/izotope-rx-12-review-2026) suggests a learned-profile pass of 6 to 10 dB in a podcast chain; treat that as a starting range, not a target. 4. **Choose Quality for the job.** A has the lowest latency and suits real time, and B can run in real time on some machines. C and D are not recommended for real-time use, so keep them for offline processing. 5. **Balance Artifact Control.** Lower values rely on spectral subtraction, which separates noise more accurately but can sound chirpy or watery. Higher values lean on wide-band gating, which leaves fewer musical artifacts but lets noise burst through right after the signal drops. 6. **Change the Algorithm only if chirps remain.** In Advanced Settings, Simple is fast and real-time capable; Advanced and Extreme leave fewer musical-noise artifacts at higher latency and CPU cost. 7. **Leave Masking at 10**, also in Advanced Settings, unless you need to cut very high, inaudible frequencies. 8. **Learn again after an FFT size change**, as the documentation advises. 9. **Shape the Reduction Curve last.** Up to 25 points let you take more out of one frequency range and less out of another. Spectral De-noise also has an Adaptive mode for noise that never settles, but it uses a significant amount of memory and CPU. iZotope points to the Adaptive mode in Voice De-noise as the more efficient option, which is one more reason to try Voice De-noise first on a moving floor. | Control | Start here | Move it when | What overdoing it sounds like | | ---------------- | ------------------------- | ---------------------------------------------------- | --------------------------------------------------------------------------- | | Threshold | 0 dB | Raise it when the noise level changes over time | Too high: quiet detail disappears. Too low: the noise pumps with the voice. | | Reduction | Low, then raise gradually | The noise still pulls attention | A thinner voice with dulled low-level detail | | Quality | A or B while auditioning | Switch to C or D for the offline pass | Higher CPU load and latency; C and D are not for real time | | Artifact Control | Leave the default | Raise for chirps, lower for noise bursts after words | Chirpy, watery tails at one extreme; gated, bursting noise at the other | ![RX 12 editor with a noise-only pause selected and the Spectral De-noise panel, with Learn, Threshold at 0 dB, Reduction and Artifact Control numbered 1 to 4 and Quality options A to D](https://storage.ghost.io/c/af/4b/af4b05d4-2e0b-4edc-8378-1e2a62fecc70/content/images/2026/10/spectral-de-noise-learn.webp) Learn from the cleanest pause you have, then work down the panel. In the noise graph, orange is the learned profile, yellow the residual noise and blue the reduction curve. ## Dialogue Isolate for traffic, crowds and changing noise Dialogue Isolate separates speech from noise that does not hold still, such as crowds, traffic, footsteps and weather, as well as from constant hiss. It uses a machine-learning model to split the audio into dialogue, noise and reverb, and you set the level of each part separately. It ships in Standard and Advanced as both a module and a plug-in, and [MusicTech's RX 12 review](https://musictech.com/reviews/plug-ins/izotope-rx-12-review/) confirms that Standard includes it. Sensitivity decides how much of the audio counts as noise or reverb. Higher values give stronger reduction but can bring artifacts and cost dialogue clarity. My advice is to raise it in small steps and stop as soon as consonants start to soften. Two refinements are Advanced-only: the Quality modes (Good/Real-time and the lower-artifact Best/Offline) and the four-band Low, Low Mid, High Mid and High controls. RX 12 improved this module. MusicTech found that Dialogue Isolate now delivers markedly higher quality than before. MixingGPT credits improved neural networks, but notes that previewing Dialogue Isolate in real time on a long file can tax even a modern machine, so audition on a short section first. Presets, the multiband split and the remaining controls are covered in the [Dialogue Isolate guide](https://soundforgepro.com/izotope-rx-dialogue-isolate/). ## Hum, buzz and one-off noises: De-hum and Spectral Repair Hum is pitched, so it needs a pitched tool. The [De-hum documentation](https://docs.izotope.com/rx12/en/de-hum.html) describes Static mode for ground noise or a simple hum with a few harmonics that do not overlap speech. Dynamic mode handles complex hum or buzz with more than a dozen harmonics or inharmonic tones. In the plug-in, Adaptive Dynamic mode is not optimized for real time, so process it offline. De-hum is in every edition, as a plug-in only in Elements. One-off sounds are a separate job. Spectral Repair's Attenuate mode lowers the spectrogram inside a selection to match the area around it, so a door slam or chair squeak blends into the background noise. The [Spectral Repair documentation](https://docs.izotope.com/rx12/en/spectral-repair.html) lists it as an editor module in Standard and Advanced, with no plug-in version. ProVideo Coalition showed how precise it can be: on camera focus noise around 500 Hz, with lower settings on every tool, Spectral Repair was the only tool in the comparison that removed the noise completely. When Attenuate is not enough, my RX Spectral Repair guide covers the Replace, Pattern and Partials + Noise modes and how long a selection each one can handle. If you work in Sound Forge Pro as well, [removing hiss and hum in Sound Forge](https://soundforgepro.com/remove-hiss-hum-sound-forge/) covers the same problem with Sound Forge's own tools. ## Check what you are removing before you commit RX 12 makes this easy. [Sound On Sound's RX12 Advanced review](https://www.soundonsound.com/reviews/izotope-rx12-advanced) describes a new ear button beside Preview, Bypass and Compare in many module windows; it switches monitoring to the signal being removed. MusicTech puts it as difference monitoring in the majority of processing modules. Use it on every pass. The difference signal should contain only noise: hiss, hum, the room. If you hear consonants, breaths or the pitch of the voice in it, the process is taking signal with it. Back off Reduction in the de-noisers or Sensitivity in Dialogue Isolate, listen again, and only then compare the result with the original using Bypass. ## Fixing watery, chirpy or pumping results Most noise-reduction artifacts trace back to one setting pushed too far. Match the symptom to its likely cause before changing anything else. | Symptom | Likely cause | Fix | | ----------------------------------------- | --------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------- | | Chirpy or watery "musical noise" | Low Artifact Control in Spectral De-noise, the Surgical filter in Voice De-noise, or the Simple algorithm | Raise Artifact Control, switch to Gentle, or try Advanced or Extreme | | Noise bursting through right after words | High Artifact Control, which behaves more like a wide-band gate | Lower Artifact Control a step | | Dull, thin or hollow voice | Too much Reduction, or high Dialogue Isolate Sensitivity | Back off, then confirm in the difference signal that the voice has gone from it | | Noise that rises and falls with the voice | Spectral De-noise Threshold set too low | Bring Threshold back toward 0 dB | | An over-processed sound overall | One heavy pass doing all the work | Use a couple of gentle passes of the same process instead | The last fix comes from [iZotope's article on dialogue background noise](https://www.izotope.com/community/blog/how-to-remove-background-noise-from-dialogue-recordings), which finds a couple of gentle passes more transparent than one extreme pass. ## Test results: how RX noise removal compares in published listening tests This composite result draws on four published listening comparisons from 2023 to 2025, run with RX 11 or an unstated earlier version. They report placings and votes rather than a shared score, so the composite is a count of outcomes, not an average. Across 13 comparisons, RX placed in the top three in 9 of the 10 where a placing was published. It won 2, both in ProVideo Coalition's RX 11 shootout, and none of the eight Production Expert tests. In the one academic test, on stationary noise, RX Spectral De-noise scored a median of 70 against 75 for the two leaders. For a wider roundup that goes beyond RX, see [audio noise reduction software](https://soundforgepro.com/audio-noise-reduction-software/). | Source | Date | RX tool and version | Field | RX result | | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------ | --------------------------------------- | -------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | [Production Expert's 2025 dialogue noise-reduction shootout](https://www.production-expert.com/production-expert-1/dialogue-noise-reduction-shootout-2025-the-results) | June 2025 | Dialogue Isolate, version not stated | 8 plug-ins, 4 examples, settings tuned by ear for each tool | 0 wins; placings 2nd, 3rd, 3rd and 4th | | [Production Expert's blind listening test results](https://www.production-expert.com/production-expert-1/results-dialogue-restoration-listening-test-noise-reduction) | January 2024 | Dialogue Isolate, version not stated | 6 plug-ins, 4 tests, blind reader vote | Hum: a near-tie, with Acon Extract: Dialogue just ahead. Roadside traffic: 3rd, as Hush Pro took 38% of votes. Total: 369 of 2,328 votes (15.9%), 3rd of 6 behind dxRevive Pro (685) and Hush Pro (636) | | [ProVideo Coalition's RX 11 shootout](https://www.provideocoalition.com/izotope-rx-11-shootout/) | May 2024 | RX 11 Dialogue Isolate; Spectral Repair | Single reviewer, 4 tests against Clear, Clarity Vx, dxRevive, Enhance Speech and RX 10 | Test 1: the reviewer's favorite. Phone voice-over: joint 2nd behind Adobe Enhance Speech. Heavy traffic plus reverb: Waves Clarity Vx won. Camera focus noise: Spectral Repair the only tool to remove it completely | | [The listening study by Steinmetz and colleagues](https://arxiv.org/abs/2310.11364) (AES) | October 2023 | Spectral De-noise, version not stated | Stationary noise, 11 recordings, 18 listeners, noisy reference scored 50 | Median 70, against 75 for Tape It and HDemucs; the gap to Tape It was not statistically significant | One condition frames all of this: every comparison above predates RX 12, whose reviews began appearing in April 2026\. The RX 12 reviews describe Dialogue Isolate as improved, with MusicTech calling its quality markedly higher, but those are reviewer verdicts rather than measurements. The 15.9% vote share is calculated from Production Expert's six published totals. My verdict: RX is the dependable toolkit, not the automatic winner. Other tools came out ahead in 11 of these 13 comparisons, yet RX finished in the top three in 9 of the 10 where it was placed. Its real strength is breadth. The same package also holds Spectral Repair, which was the only tool to clear the camera noise in ProVideo Coalition's test, plus De-hum and learned-profile Spectral De-noise for steady floors. ProVideo Coalition reached a similar conclusion: with RX 11, iZotope caught up with Supertone Clear and Waves Clarity Vx and was perhaps more reliable, with a lower overall artifact level. For heavy, changing noise behind dialogue, try Dialogue Isolate first and keep an alternative in mind. The [best iZotope RX alternatives](https://soundforgepro.com/best-izotope-rx-alternatives/) covers the tools that beat RX in specific tests, and [CrumplePop vs iZotope RX](https://soundforgepro.com/crumplepop-vs-izotope-rx/) compares a simpler plug-in route. ## Which RX 12 edition you need for noise removal For background noise, the edition decides which of these tools you get. [iZotope's RX Elements page](https://www.izotope.com/products/rx-elements) lists six plug-ins (De-click, De-clip, De-hum, De-reverb, Voice De-noise and Repair Assistant) and no standalone editor. The editor comes with Standard and Advanced. | Tool | Elements | Standard | Advanced | | ---------------------------------- | -------- | ------------------ | ------------------------------------------------------------- | | RX Audio Editor | No | Yes | Yes | | Voice De-noise | Plug-in | Module and plug-in | Module and plug-in | | De-hum | Plug-in | Module and plug-in | Module and plug-in | | Spectral De-noise | No | Module and plug-in | Module and plug-in | | Dialogue Isolate | No | Module and plug-in | Module and plug-in, plus Quality modes and multiband controls | | Spectral Repair | No | Editor module | Editor module | | De-wind | No | No | Editor module | | Ambience Match | No | No | Editor module; plug-in in AudioSuite only | | US store price on October 20, 2026 | $99 | $399 | $1,399 | The prices come from iZotope's US store on October 20, 2026, as listed on the [RX Standard page](https://www.izotope.com/products/rx-standard) and its sister pages; confirm the figure at checkout, because store prices change. My pick for background-noise work is Standard. It adds the editor, Spectral De-noise, Dialogue Isolate and Spectral Repair, which together cover steady floors, changing noise behind speech and one-off events. Elements makes sense if Voice De-noise and De-hum inside your DAW are all you need. Advanced earns its higher price only if you need De-wind, Ambience Match, or Dialogue Isolate's Best/Offline quality and multiband controls; the [RX Advanced page](https://www.izotope.com/products/rx-advanced) lists its extra modules. If you are choosing an edition for more than noise work, the [RX 12 Elements vs Standard vs Advanced comparison](https://soundforgepro.com/izotope-rx-elements-vs-standard-vs-advanced/) covers every module. The [iZotope RX hub](https://soundforgepro.com/izotope-rx/) groups the other repair guides by job. ## When RX cannot rescue a recording - **Heavy wind.** The [De-wind documentation](https://docs.izotope.com/rx12/en/de-wind.html) describes a tool for light to moderate wind bursts, not for wind that blows out and distorts the microphone signal. For other approaches to wind on a voice track, see [reducing wind noise in Sound Forge](https://soundforgepro.com/reduce-wind-noise-voice-recording-sound-forge/). - **Very deep reduction.** Strong suppression degrades low-level signal along with the noise, so a very noisy recording can come out cleaner but thinner. Aim for less objectionable, not silent. - **Aggressive isolation.** High Dialogue Isolate Sensitivity buys stronger reduction at the cost of artifacts and dialogue clarity. - **Ambience that no longer matches between edits.** [Ambience Match](https://docs.izotope.com/rx12/en/ambience-match.html) in Static mode is meant for unchanging ambience such as room tone, air conditioners or fans, and it is an Advanced-only tool. Gaps of dead silence between edits? See [room tone in Sound Forge](https://soundforgepro.com/room-tone-sound-forge/). ## Doing the same job in Sound Forge Pro Sound Forge Pro has its own route, explained in [how to remove background noise in Sound Forge Pro](https://soundforgepro.com/how-to-remove-background-noise-in-sound-forge-pro/). If you are deciding between the two editors rather than following steps, [Sound Forge Pro vs iZotope RX](https://soundforgepro.com/sound-forge-pro-vs-izotope-rx/) compares them side by side. ## Frequently asked questions ### Can RX Elements remove background noise? Yes, within limits. Elements gives you Voice De-noise for steady floors and De-hum for hum, both as plug-ins inside your DAW or editor. Spectral De-noise, Dialogue Isolate, Spectral Repair and the standalone editor need Standard or Advanced. ### Does Adaptive mode replace Learn? Not entirely. Adaptive mode in Voice De-noise follows a changing floor but sets its threshold lower to prevent artifacts, so its result can differ from a Learned threshold. Spectral De-noise's Adaptive mode costs a lot of memory and CPU. When you have a clean stretch of noise, I would Learn from it; Adaptive is the fallback when there is none, or when the noise keeps moving. ### Should I remove hum before broadband noise? I would start with De-hum. It targets pitched tones and their harmonics, and once those are gone, Spectral De-noise can learn the remaining floor without the hum shaping the profile. Check the difference signal after each step, as with any pass. ### What are RX 12's system requirements? MusicTech's RX 12 review lists macOS Sonoma 14.7.x or later, and Windows 10 22H2 or Windows 11 24H2.