iZotope RX 12 De-bleed: Headphone and Mic Spill

Use RX 12 De-bleed to reduce headphone leakage or instrument spill while keeping the performance you recorded. Choose Classic in the Audio Editor when you have a clean, aligned copy of the leaking source. Choose the reference-free mode when the wanted source matches one of its instrument categories. The new real-time plugin uses that reference-free approach; Classic remains an Audio Editor workflow.

Start with modest reduction and listen in the mix. A quieter leak is useful only if the vocal, guitar or drum still sounds convincing. New hollow, watery or phasey coloration is a reason to compare against the original and back off, even when the unwanted sound has almost disappeared.

What counts as bleed

Bleed is sound captured by a microphone intended mainly for another source: click track in a vocal mic, guitar in a vocal take, snare in a tom mic or cymbals across the drum kit. You may have a separate recording of the leaking source, but that is not guaranteed.

That is different from steady room noise, hum or reverb. Use De-hum for a fixed harmonic ladder, the background-noise workflow for hiss or fans, and De-reverb for reflected room tails.

Not all spill is bad. Drum-mic bleed can carry the kit's spatial coherence, and instrument spill may make a live performance feel connected. Process only when the leakage creates a concrete mix problem.

RX 12 changed the De-bleed decision

Older De-bleed tutorials describe one method: provide a separate source track, align it with the contaminated track, Learn the relationship and reduce it. RX 12 retains that as Classic mode but adds reference-free machine-learning separation and a real-time plug-in.

The official RX 12 upgrade overview introduces the rebuilt De-bleed. The current De-bleed manual documents both the reference-free controls and Classic workflow. De-bleed is available in RX 12 Standard and Advanced as a module and plugin.

A clean reference remains worth testing when it matches the leak. It gives Classic a recorded source to compare with the active track. The reference-free mode is useful when that source is unavailable and its instrument categories fit the recording.

A 2026 discussion about a supposedly missing reference option illustrates the confusion: the account named iZotopeOfficial points users to the Classic button in the editor and says Classic is absent from the new plugin. The current manual independently documents the retained Classic workflow. A missing reference selector in the plugin does not mean the editor lost it.

Classic versus reference-free De-bleed

If both are possible, test them on the same difficult phrase. Choose the result that preserves the wanted source, not the workflow with the newer label.

Nine steps for safe bleed removal

1. Preserve and diagnose. Work on an independent copy of the audio and confirm that the leakage causes a problem in the mix.
2. Choose the mode. Use Classic in the Audio Editor with a valid reference, or choose the wanted Signal Type in reference-free mode.
3. Prepare the input. For Classic, open the source and active files, match sample rates and align their events. For reference-free mode, choose a revealing region of the target recording.
4. Learn or select. In Classic, select obvious bleed in the active track and click Learn. Reference-free mode uses its Signal Type instead of a reference-track Learn pass.
5. Adjust the amount. Start with modest Reduction strength in Classic or a modest De-bleed amount in reference-free mode; raise it only while the wanted source remains intact.
6. Check artifacts. In Classic, balance Artifact smoothing against dullness. In reference-free mode, reconsider the amount, selected source and processing frequency range.
7. Compare fairly. Audition alternatives from the same unprocessed input at matched loudness. Use host bypass for the plugin and Compare where available in the editor.
8. Inspect what is kept. Check attacks, tone and tails. In reference-free mode, audition Invert bleed, then turn it off and judge the normal processed output.
9. Approve and export. Listen through the affected take with related microphones active, apply the chosen processing once and export to a new filename while preserving the original.

Use the editor's Compare controls where available to audition candidates. If testing by rendering, undo all renders from the current De-bleed experiment before testing the next candidate; retain any earlier approved repair. Do not compare a second treatment of already-processed audio against a first treatment of the original. The selection, Preview, Compare and History guide explains how to keep that comparison fair.

Prepare a Classic reference correctly

The active track contains wanted audio plus bleed. The bleed source track should contain the leaking signal by itself. For headphone spill, that may be the exact cue mix; for a click, the original click track.

The current Classic documentation requires two open files, matching sample rates and alignment within a few milliseconds. The same audio event must occur at the same timeline position in both. If sample rates differ, use RX's Resample module on a working copy as needed, then verify timing; changing only a rate label is not the same as resampling.

Check alignment near the beginning, middle and end of the section. Different edit points, time stretching or a cue exported from a different arrangement can break the match. Proper sample-rate conversion is not itself a reason for drift. Do not move the wanted performance out of sync with the DAW merely to fit the wrong reference.

Learn a representative bleed relationship

Open both files in the RX Audio Editor and switch the De-bleed module to Classic. Keep Composite View off: the file manual does not recommend De-bleed there. Confirm that Active track names the wanted-plus-bleed recording. Choose the clean leaking source in Select bleed source track, then select a passage in the active file where the bleed is most obvious.

Click Learn. The manual asks for obvious bleed; it does not require wanted audio to be present or say that a bleed-only passage creates a worse model. Choose a section that reveals the leak rather than an arbitrary fixed duration. In the learned profile display, switch between Bleed source and Active track to check that the corresponding events are present and aligned.

After Learn, select the part you want to treat and audition a short region. Learning and processing selections can differ. Use Render to apply the treatment in RX 12. If the result is poor, check the selected source and alignment before raising Reduction strength or processing the whole take.

Set Classic Reduction strength

In Classic mode, Reduction strength controls the amount of bleed reduction. iZotope recommends beginning low because higher settings can remove wanted audio. A valid source relationship improves the starting conditions; it does not guarantee complete, transparent removal.

Raise it until the spill no longer pulls attention in the full mix, then check the wanted track alone. Stop if vocal formants thin, guitar tone swirls or drum attacks lose body. Solo can reveal damage; the arrangement decides whether the remaining bleed matters.

In reference-free mode, the amount control is the De-bleed knob. Gain adjusts the processed output level and can help match perceived loudness for comparison. These controls are separate from Classic's Reduction strength and Artifact smoothing; do not search for a universal setting that fits every recording.

Artifact Smoothing: watery versus dull

Classic De-bleed uses FFT-based analysis. Aggressive removal can create “musical noise,” often heard as underwater chirps or a moving metallic texture. Artifact Smoothing can reduce that residue.

Increase smoothing only until the watery texture settles. Too much can dull the wanted track. If you need extreme smoothing to tolerate the result, lower Strength or relearn from a better-aligned section.

Check quiet phrase endings and cymbal decays. Loud passages hide processing artifacts that become obvious in exposed gaps.

Reference-free De-bleed for instruments and drums

Choose the signal you want to isolate in Signal Type. The eight categories are Vocals, Snare, Kick, Toms, Cymbals, Drums, Guitar and Piano. For hi-hat leaking into a snare microphone, start with Snare. For drums leaking into acoustic guitar, start with Guitar. Choosing the unwanted instrument would ask the processor to keep the wrong source in its normal output.

Raise the De-bleed amount gradually. Use the Frequency Range handles to confine processing to a useful band when full-range treatment damages the wanted source. The display overlays the processed spectrum in front of the original spectrum during preview; the picture helps locate a change, but listening decides whether it is useful.

Invert bleed switches the output to isolate the bleed signals. Audition it as a diagnostic, then turn it off before judging or exporting the wanted track. Hearing some wanted material in the inverted output is a reason to inspect the normal output carefully, not a precise measurement of damage. Classic's documented controls do not include this button.

There is no Bass category in this selector. For a broad Voice/Bass/Drums/Other split, test Music Rebalance instead. If the wanted source does not suit the De-bleed categories and no valid Classic reference exists, manual edits or leaving some spill may be the better result.

Check drum processing with all microphones playing. A tom mic that sounds isolated on its own may weaken the snare image or change the combined cymbal sound. Listen for changes in low end, ambience and stereo position.

The current RX edition comparison lists De-bleed for Standard and Advanced, not Elements. Both the full Audio Editor and this module are outside RX 12 Elements. Edition choice does not remove the acoustic problem: overlapping sources can still leave residue or lose detail.

Use the real-time plugin in your DAW

The reference-free plugin lets you audition reduction while the mix runs. Use the DAW's bypass to compare it at matched loudness. If you need Classic's separate reference file and Learn workflow, open the files in the Audio Editor rather than looking for those controls on the plugin.

You do not have to test both plugin and module for every repair. Use the route that fits the job, then listen to the actual printed or exported result. Keep an independent original audio file; duplicating a DAW track may still leave both tracks pointing to the same source file.

Watch CPU and latency in a dense session. Freeze or commit only after the setting passes, and keep the original track available.

For editor work, the RX file workflow distinguishes an RX Document, which retains editable history, from exported audio for the DAW. Export the approved audio to a new filename and check it in the session. Saving an open WAV or AIFF can overwrite its original file, so preserve the source before making destructive saves.

Headphone bleed in a vocal

First confirm that the cue mix or click is audible in the actual vocal mix. If it disappears under production, processing may do more damage than the leak.

When the exact headphone source exists, Classic mode is a strong first test. Align it to the microphone track before vocal comping changes timing. Process short sections separately when edits break the relationship.

Protect consonants, breaths and pitch. A stronger setting that exposes the cue less but makes the singer hollow is not an improvement. The RX vocal-cleanup workflow helps place bleed repair alongside clicks, plosives, hum and room problems without requiring every module on every take.

Drum and multi-mic spill

Decide what you want to change: reduce hi-hat in the snare mic, isolate tom attacks, or calm cymbal wash. “Remove all bleed” is not a mix objective.

When no valid Classic source is available, choose the relevant reference-free category and test a section containing quiet hits, accents and decays. Compare that microphone inside the full kit, with the normal bus compression active. Check the stereo image and a mono sum: changing the contribution of one microphone can change the combined sound even if the soloed track seems cleaner.

The broader Sound On Sound guide to microphone bleed explains why mic pattern, placement, phase and arrangement matter before software. RX can reduce a capture problem; it cannot replace coherent tracking.

De-bleed versus a gate, expander or stem separator

A gate reduces the whole signal according to its detector level, normally the input or a selected sidechain. It does not automatically recognize the wanted instrument. As Waves explains in its gating guide, the threshold and timing must preserve quiet wanted hits while rejecting spill. Once the gate opens, overlapping bleed still passes. A downward expander makes the level reduction more gradual, but does not separate simultaneous sources.

Classic De-bleed learns a reference relationship; reference-free De-bleed isolates the selected instrument category. Music Rebalance separates a music mix into Voice, Bass, Drums and Other. It can be worth testing for broad drum-versus-vocal spill, but it is not the same as supplying Classic with an exact leaking source.

Dialogue Isolate separates voice, noise and reverb for spoken dialogue. Use it when speech clarity is the target, rather than assuming it is a general instrument separator. Choose a processor for the actual problem and compare alternatives before stacking separation stages.

Why De-bleed sounds worse

Watery or metallic: compare with the original to confirm the coloration is new. Reduce the amount first. In Classic, increase Artifact smoothing only enough to reduce musical noise; that control does not belong to the reference-free interface.

Dull: in Classic, try less Artifact smoothing or Reduction strength. In reference-free mode, lower De-bleed or narrow the processing range. Check that the wanted source is selected.

Flanging or phase shift: check Classic source alignment and compare with the untreated recording. In a multi-mic session, listen to how the processed microphone combines with the others before changing their timing or polarity.

Bleed remains in some sections: check for edits, source changes or headphone movement that may alter the Classic relationship. Realign and relearn the affected section when appropriate. For reference-free processing, retest the source category and amount on those passages.

Wanted performance pumps: lower the amount and compare the same input again. Check for existing compression or gates as well as separation artifacts. The sound alone does not prove which internal model behavior caused it.

Solo is clean but mix is weak: bypass the treatment with all related microphones active. Useful ambience or shared instrument detail may have been lost; keep more spill if the combined result is stronger.

Prevent bleed before RX

Use well-sealing closed-back headphones for tracking and reduce the cue level to what the performer needs. Keep both earcups properly seated. Try microphone placement and rejection angles that reduce spill without spoiling the wanted tone; the appropriate null depends on the microphone's polar pattern.

For multi-mic instruments, placement, null direction, phase checks and arrangement reduce the amount software must separate. Record a clean click or cue source when Classic De-bleed may be needed later.

Before recording the full take, listen to a short test with the cue playing. Confirm that the performer can still pitch and keep time comfortably after any monitoring change. Keep the exact cue or click file if you may need a Classic reference later.

Final De-bleed quality gate

• The spill causes a specific problem in the full mix.
• Classic is used only with a clean, matching, time-aligned reference.
• Reference-free mode uses the wanted Signal Type, and Invert bleed is off for the wanted output.
• Reduction stops before wanted tone, attacks or room disappear.
• No watery, metallic, dull or phasey artifact is introduced.
• Drum and multi-mic results are checked with all related tracks active.
• The chosen route and actual exported result are compared with the original at matched loudness.
• The untouched recording remains available.

For adjacent vocal, instrument and stem workflows, continue through the Music & Vocals hub.

Frequently asked questions

Does RX 12 De-bleed need a reference track?

The reference-free mode does not. Choose the wanted Signal Type instead. Classic in the Audio Editor still requires a clean, aligned bleed source; Classic is not available in the new plugin.

Is De-bleed in RX 12 Standard?

Yes. RX 12 Standard and Advanced include the De-bleed module and real-time plugin. RX 12 Elements does not include De-bleed or the standalone Audio Editor.

Can De-bleed remove headphone music from vocals?

It can reduce headphone leakage, especially when the exact cue mix is available for Classic. Match sample rates and alignment. Complete removal is not guaranteed, and leaving some residue may preserve the vocal better.

Can RX remove drum microphone bleed?

RX 12 reference-free De-bleed includes Snare, Kick, Toms, Cymbals and Drums categories. Select the source you want to keep and check the result in the full kit, including a mono comparison.

How closely must a Classic reference align?

The manual calls for matching sample rates and alignment within a few milliseconds, with corresponding events at the same timeline positions. Check the beginning, middle and end of the processed section.

Why does De-bleed sound underwater?

Aggressive reduction can produce watery artifacts. Lower the amount first. In Classic, increase Artifact smoothing if needed and reduce it if the audio becomes dull. Reference-free mode instead has De-bleed amount and Frequency Range controls.

Should I use De-bleed or a gate?

Try a gate or downward expander for spill between wanted events, while preserving quiet hits and tails. An open gate still passes overlapping bleed. De-bleed is worth testing when simultaneous sources need separation.

Should I process the whole track?

Only after testing the different conditions in the take. Classic may need realignment and a new Learn pass across edited or changing sections. Retest reference-free settings where the arrangement changes, then check the full result before export.