iZotope RX 12 De-rustle: Remove Lav Mic Noise

Use iZotope RX 12 De-rustle when a lavalier microphone has captured clothing rubbing, brushing or soft body-movement thuds. In RX 12 it is an Advanced-only module, with an AudioSuite-only plug-in version. Start with low Reduction Strength, preserve enough ambience to keep the room stable, and process short affected passages when the wardrobe or movement changes. Stop when consonants thin out—even if a little rustle remains.

De-rustle is not a general “bad lav” button. Wind, plosives, cable knocks, electrical hum, clipping and room noise need different routes. The fastest clean repair comes from identifying the defect first, testing the worst representative line, and using Spectral Repair only for the isolated impacts the separation cannot remove transparently.

De-rustle at a glance

This diagnosis matters because clothing rustle overlaps speech in both time and frequency. A broad low-pass, high-pass or midrange cut cannot know which energy belongs to the actor and which belongs to a shirt. The 2018 DAFx paper on neural lavalier-noise separation explains that overlap and the difficulty of tracking unpredictable friction. It provides research background, not a specification of RX 12’s current network or a promise of complete recovery.

What RX 12 De-rustle actually does

The current RX 12 De-rustle manual describes a machine-learning system trained with isolated rustle, clean dialogue and dialogue mixed with rustle. It attempts to separate the interference from the voice rather than merely turning down a fixed frequency band. That makes it appropriate for lav capsules brushing fabric as the wearer moves.

The same documentation treats rustle as a range, from high-frequency brushing to low-frequency thuds. Those components do not always respond equally. A setting that clears the scratchy layer may leave one chest-level impact; pushing the global strength until that last thud disappears can damage an entire sentence. Think in two stages: broad separation for the recurring bed, then local repair for exceptions.

De-rustle is currently included in RX 12 Advanced. In the standalone editor it is a module; the plug-in form is AudioSuite only. Do not plan a universal live insert workflow around it. For DAW handoff decisions, our RX and DaVinci Resolve/Fairlight guide shows when to send a clip out, render, and return it to the edit.

Diagnose clothing rustle before opening the module

Play the problem at normal speed first. True clothing rustle usually changes with an actor's shoulder, torso, arm or wardrobe movement. In the spectrogram it may look like irregular wideband scratches, fuzzy bursts or lower-frequency impacts rather than one continuous horizontal line. Compare a stationary sentence with a movement-heavy sentence. If the noise follows “p” sounds instead of body movement, test De-plosive. If it continues unchanged while nobody moves, test a denoise module.

Check the lav track against the boom, camera scratch track and any second transmitter. A clean boom phrase can replace a badly damaged lav word more naturally than extreme processing. If both microphones contain the event, the source may be movement in the room, a prop or wind rather than the concealed capsule. Keep every alternate until the repaired edit has passed lip-sync and room-tone review.

For background-noise ambiguity, use the decision framework in Voice De-noise versus Spectral De-noise. Use De-rustle for the clothing contact, then assess any remaining noise on its own. If an essential word is buried beyond recognition, preserve the original and seek an alternate take or replacement dialogue rather than guessing what was said.

Prevent lav rustle at the source whenever you still can

Post repair is the fallback, not the first line of defense. Move the capsule away from seams, necklaces, zippers and loose layers. Build a small air gap so fabric slides around the mount instead of across the grille. Secure the cable with a relaxed strain-relief loop; a tightly pulled cable can transmit each body movement directly into the capsule.

Rycote’s Undercovers combine soft fabric discs with Stickies to reduce clothing rustle around concealed microphones. Bubblebee's guide to hiding lavs without unwanted noise explains how fabric guards create space, rubber mounts absorb friction and a cable saver supplies strain relief. Moleskin, fur and purpose-built mounts can all work, but the right choice depends on fabric, perspiration, movement and how visible the mount may be.

Record a movement test before the take: arms up, torso twist, sitting, standing and the loudest costume motion in the scene. Monitor the actual lav, not only the boom. In a March 2024 discussion about removing shirt rustle, contributors suggested several different repair tools, while two replies emphasized fixing the recording on location. Those are individual practitioner suggestions, not an RX 12 test or a consensus that one processor always works.

Protect the source and build a test region

Duplicate the production file before processing. Preserve its sample rate, channel count and timing, and retain the original timecode metadata for the handoff. In a DAW, keep an untouched duplicate aligned with the repair and solo one version at a time; playing both together changes the comparison. In the RX editor, use History or a separate original file for rendered-edit comparisons. A module’s Bypass applies during Preview; it does not undo processing already rendered into the file.

As a practical starting point, select roughly fifteen to thirty seconds containing ordinary dialogue, a quiet phrase, a typical brushing passage and some natural ambience. This is an audition example, not a required duration or a processing limit. Include hard consonants and word endings. If the scene contains both a silk blouse and a jacket, treat them as different sources; the friction spectrum and movement pattern may change enough that one setting is misleading.

Use consistent pre-roll and monitoring level for every Preview. The RX selections, Preview, Compare and History guide covers this audition discipline in detail. A repair that only sounds impressive because it is louder has not passed.

Set Reduction Strength conservatively

Reduction Strength is the main tradeoff. The manual states that lower values leave more rustle but preserve more dialogue clarity; higher values remove more rustle but can reduce clarity. Start low. Raise the control until the clothing stops pulling attention away from the line, then listen specifically to “s,” “f,” “t,” breath detail and the tail of quiet words.

There is no defensible universal percentage. A close, bright lav under a cotton shirt is not the same as a capsule touching a synthetic jacket during a fight scene. The useful stopping point is perceptual: the line becomes easier to follow, the actor still sounds like the same person, and the remaining texture sits under the scene rather than announcing itself.

If one brief scrape forces a setting that hollows the rest of the take, lower the global strength and mark the scrape for a local second pass. Keep that second pass confined to the remaining event and compare the cumulative result with the original. Rendering is reversible through retained History; losing the original or overwriting the only usable copy is the avoidable risk.

Balance Ambience Preservation without pumping the room

Ambience Preservation controls how much background stays in the result. This is not merely a taste knob: the De-rustle algorithm may classify some real room ambience as rustle. Too little retained ambience can make the voice jump into a vacuum during movement; too much can return part of the clothing texture.

Loop the transition into and out of a rustle event. Listen for a room that collapses on one word, swells between phrases or changes color every time the actor moves. Judge the line in the production mix as well as solo. A little steady background is often less conspicuous than a perfectly clean phrase surrounded by pumping.

If the repair leaves a hole, fill it from clean room tone only after the dialogue itself is stable. Do not use ambience as camouflage for damaged consonants. When the background is changing traffic, crowd or weather rather than clothing, a dedicated dialogue-separation route may be more suitable.

Choose the correct channel-processing mode

Channel Independent processes each channel separately and is the fastest mode. It is the sensible default for a mono lav and for efficient Preview. The manual describes it as the efficient real-time preview/processing option, but that does not change the AudioSuite-only limitation of the plug-in version.

Joint Channel is described as a higher-quality separation option, particularly for stereo files with similar content in both channels. “Particularly” does not mean stereo is its only permitted use. Compare it on the actual recording instead of ruling it out for mono. Treat two independent receivers or a lav/boom split as separate sources; a two-channel file is not automatically a stereo image.

Advanced Joint Channel adds further processing and is the vendor’s highest-quality option, especially at high sample rates, with longer processing times. Test it on the same source and selection after the basic controls work. Use Compare to audition candidates when evaluating a costly setting, and keep the version that preserves speech and room continuity. Do not upsample a recording just to fit that description or assume a slower render guarantees a better result.

Process local regions instead of forcing one setting across the take

Rustle is unpredictable because clothing, pressure, posture and movement keep changing. Split the edit by acoustic condition: seated interview, walking section, arms raised, jacket added, transmitter moved. Give each condition a short verified setting, and leave clean regions untouched.

Extend selections slightly beyond the obvious visible event so the repair does not switch abruptly. Preview the boundary in context. If the restored passage changes tone at the selection edge, adjust the region, use a gentler pass or crossfade a rendered duplicate. Keep a log of timestamps and settings when a long documentary scene requires many repairs.

This local approach also makes review faster. A director or dialogue editor can compare twenty marked events instead of trying to hear subtle damage across a forty-minute blanket render.

Repair stubborn thuds and hard scrapes separately

After De-rustle, return to isolated body thumps, cable knocks and hard scrapes. In the spectrogram, select the event narrowly in time and frequency while protecting the speech harmonic structure. The current Spectral Repair manual supports Attenuate, Replace, Pattern and Partials+Noise routes; the right choice depends on whether you are suppressing an event or reconstructing a damaged gap.

Our Spectral Repair guide covers those modes and selection shapes. For a thud under a vowel, start with a small low-frequency selection and audition in context. Do not erase the voice's fundamental just because the impact occupies the same band. A short gain envelope or a patch from the boom may be cleaner than increasingly complex synthesis.

The 2018 DAFx research supplement illustrates speech, rustle and room tone in spectrograms and provides audio examples. It helps explain why preserving ambience matters; its examples are historical research material, not a listening test performed for this guide.

Place De-rustle carefully in the dialogue chain

Repair gross capture defects before creative EQ, compression or loudness work. A practical order is: restore clipped peaks if they are truly clipped, remove dominant clothing contact, repair individual impacts, reduce any remaining stable noise, then shape tone and dynamics. The exact order can change when one defect masks another, so test on the representative region.

Avoid heavy denoise before De-rustle if it makes the friction smear into speech. Avoid compression before repair because it can raise quiet rustle and make local impacts trigger gain reduction. After De-rustle, compare any Voice De-noise pass at conservative settings; stacked separation can sound metallic even when each module seems acceptable in isolation.

The Clean Dialogue in RX hub connects clothing-noise repair with hum, wind, plosives, mouth sounds, reverb and general dialogue workflows. Use it to route the next defect rather than adding every module to this one chain.

Step-by-step RX 12 De-rustle workflow

1. Protect the original recording. Duplicate the lav track, preserve the boom and other alternates, and keep an untouched source for comparison.
2. Confirm the sound is clothing rustle. Separate brushing and body-movement thuds from wind, plosives, cable knocks, hum, clipping and room noise.
3. Select a representative test passage. Include ordinary speech, quiet words, typical rustle and natural ambience.
4. Start with low Reduction Strength. Raise it only until the clothing movement recedes without thinning consonants or word endings.
5. Balance Ambience Preservation. Keep background continuity stable without returning the distracting friction.
6. Choose the channel algorithm. Start with Channel Independent for speed; compare Joint Channel and Advanced Joint Channel on the same source. Similar-content stereo is a documented strength of joint processing, not its only permitted input.
7. Render local passes. Divide the take when wardrobe, movement or microphone contact changes instead of forcing one aggressive setting across everything.
8. Repair isolated thuds separately. Use Spectral Repair, a short gain edit or an alternate microphone for the remaining impacts.
9. Audit speech and ambience. Level-match the result, listen to consonants and quiet word endings, check room-tone continuity and compare with the untouched source.
10. Export and verify the handoff. Retain an RX Document and separate output audio, preserve the required Broadcast WAV metadata, then reimport and check timing, channel order and the full repaired passage.

Save a revision after each successful stage. If a later denoise or EQ pass makes the voice brittle, return to the clean De-rustle checkpoint rather than stacking compensating processors. The RX file guide distinguishes Save RX Document, which retains the original and edit history, from an exported audio file for the DAW. Ordinary Save on an opened WAV or AIFF overwrites that file; do the work on copies. For a Broadcast WAV handoff, use the required BWF and Preserve Non-Audio Data options, then reimport and verify the start position, duration, channel order and lip sync. Metadata settings do not replace that check.

Quality-control checklist

• Compare at the same monitoring level from the same pre-roll.
• Listen to the quietest essential line, not only the loudest actor.
• Check “s,” “f,” “t,” “k” and breath detail for thinning.
• Watch and listen through every local-selection boundary.
• Check the repaired lav against the boom and picture sync.
• Confirm the room does not pump or vanish during movement.
• Audit the result in the full mix, where a little residue may disappear naturally.
• Keep the untouched source and a reversible repair session.

If RX becomes slow or stops responding while auditioning a long high-quality pass, use the diagnostics in RX slow or freezing before assuming the audio itself is corrupt.

What not to do

• Do not use De-rustle on every noisy lav without diagnosing the sound.
• Do not chase the last faint scrape by sacrificing dialogue clarity.
• Do not process an entire take with a setting chosen from one extreme event.
• Do not treat a two-channel container as correlated stereo without inspecting it.
• Do not overwrite production audio or discard the boom alternative.
• Do not stack aggressive De-rustle, denoise and compression without comparing the complete result with an untouched source. Module Bypass does not reverse earlier rendered repairs.
• Do not invent a universal Reduction Strength percentage.
• Do not expect post-processing to outperform a properly isolated lav at capture.

Frequently asked questions

What does iZotope RX De-rustle remove?

It targets noise made when a lavalier microphone rubs or brushes clothing, including irregular high-frequency brushing and some low-frequency thuds. It is not a general wind, hum, clipping or room-noise remover.

Is De-rustle included in RX 12 Standard?

No. The current RX 12 manual lists De-rustle as an Advanced module. Its plug-in version is AudioSuite only, while the standalone RX Audio Editor provides the module workflow.

What is the best De-rustle Reduction Strength setting?

There is no universal value. Start low on a representative passage and raise it until the rustle stops distracting. Stop when consonants, breaths or quiet word endings lose clarity, even if a little clothing texture remains.

What does Ambience Preservation do in De-rustle?

It controls how much background ambience remains because the algorithm may classify some of that background as rustle. Balance it so the room stays continuous without returning the unwanted clothing noise.

Should I use Channel Independent or Joint Channel?

Channel Independent is the fastest option. Joint Channel offers higher-quality separation, especially for similar-content stereo, but the manual does not restrict it to stereo. Advanced Joint Channel adds processing for the highest-quality option at a greater time cost. Compare on the actual source and keep unrelated microphones separate.

Can De-rustle remove wind noise?

It is not the first choice. De-rustle is trained for clothing-contact noise. Diagnose turbulent wind separately and use a wind-removal workflow; use De-plosive for short pressure bursts tied to “p” and “b” sounds.

What should I do when De-rustle leaves low-frequency thuds?

Keep the transparent De-rustle pass, then repair each remaining impact with a narrow Spectral Repair selection, short gain edit or clean alternate-microphone patch. Do not raise global strength until the whole sentence becomes hollow.

How do I prevent lavalier clothing noise?

Keep the capsule away from seams and jewelry, create space between grille and fabric, isolate it with a suitable soft or rubber mount, secure a relaxed cable strain-relief loop and record a full movement test before the take.