RX 12 Music Rebalance: Separate Vocals, Drums and Bass

RX 12 Music Rebalance is useful when you have a finished mix but no multitrack: bring a buried voice forward, reduce dominant drums, make a rehearsal instrumental, or split the file into four editable stems. It does not recover the original session. It estimates four categories—Voice, Bass, Drums and Other—and the quality of those estimates depends on how much the sounds overlap.

Start by deciding whether you need one new stereo balance or separate stem files. That choice determines whether to render the Music Rebalance module or use Stem Split and Stems View.

What Music Rebalance separates in RX 12

The current RX 12 Music Rebalance manual names four elements:

• Voice: lead and supporting vocal material the model classifies as voice.
• Bass: bass elements identified in the mix.
• Drums: drums and percussion.
• Other: everything not identified in the first three groups, typically melodic instruments.

Older tutorials and search snippets often say Vocals, Bass, Percussion and Other. For RX 12, use the labels in the current interface and manual: Voice, Bass, Drums and Other. Guitar, piano and strings do not receive their own named lanes; they usually share Other, which is why a clean “guitar-only stem” is not a promised output.

Choose Rebalance or Stem Split before processing

Use the normal Music Rebalance module when the goal is a new stereo mix: lower Voice, lift Bass, mute Drums, or make a small balance correction. The module’s gain sliders, Mute and Solo controls let you audition and then render that combined result.

Use Stem Split when you need four independent files or want to edit one separated lane with another RX module. Stem Split sends the result to a dedicated Stems tab. Crucially, the manual states that Stem Split ignores the module’s gain slider positions and assumes 0 dB for each stem. Moving Voice to −10 dB and then clicking Stem Split does not create a “quiet voice” stem; it creates separated stems at their normal split level.

The Stem Split pane guide recommends the ordinary module and Render button when you only need mute, solo or level changes. It keeps a simple level correction in the original file tab, without creating a separate Stems tab.

Prepare a revealing test section

Open a lossless copy of the stereo master when available. Prefer the original WAV or AIFF over a streaming copy that has already lost information. A lossy file can still produce a useful result; it just gives you another possible source of artifacts.

Select a difficult 20–30 second passage as a first listening test, not a required analysis length. Include vocal and drum overlap; repeat the test on a transition if it falls elsewhere. A sparse intro can hide problems that become obvious in the chorus.

Open Music Rebalance, set gains to 0 dB, choose Good / Real-time, and use Preview with one category Soloed. Repeat for each category. This auditions the separation without applying it to the file. The RX editor procedure below uses its Preview, Compare and Render controls; the AudioSuite exception is explained under quality modes.

Set Sensitivity without turning everything up

Sensitivity tells the algorithm to pay more attention to one stem. Increasing Voice Sensitivity can preserve subtle breaths and quiet vocal detail, but iZotope warns that it may also pull more unrelated sound into Voice. Listen for cymbal wash, guitar harmonics and reverb moving with the vocal.

The controls are relative. Raising Voice, Bass, Drums and Other all to 100% does not make the entire separation more sensitive; the manual says that simultaneous change produces no separation difference because the model cannot pay more attention to everything at once.

Change one Sensitivity control and Preview the same Soloed stem again, or add a Compare candidate from the still-unprocessed selection. If detail returns with acceptable leakage, keep it. If the stem becomes fuller only because other instruments came with it, move back. Do not repeatedly Render new tests over an already processed excerpt.

Good, Better and Best quality modes

The current Quality menu offers three modes:

Start with Good to judge feasibility, then compare Better and Best on the same difficult selection. Keep the same stem Soloed, gain and Sensitivity settings when comparing quality alone. With all four stems combined at 0 dB, reconstruction can hide differences between the individual separations. Best is the documented highest-quality option; listen for whether it actually improves the stem you need.

In Pro Tools, note that the RX 12 Music Rebalance AudioSuite plug-in does not provide Preview before rendering. Test a duplicate clip or short selection and compare the rendered result rather than assuming the real-time audition workflow applies.

Use Solo and Mute as an artifact audit

Solo every stem, even if you only care about Voice. Separation errors often reveal themselves in the stem you plan to turn down:

• Voice: listen for cymbals, snare brightness, guitar sustain and reverb.
• Bass: listen for kick attack and low synth material moving together.
• Drums: listen for piano or guitar transients softened with percussion.
• Other: listen for vocal ambience, consonants or bass harmonics left behind.

Clear Solo, then Mute the target to hear what remains. If removing Voice tears holes in instruments, full vocal removal is too destructive for that source. Modest attenuation may still help. Clear every audition Solo and Mute before preparing a normal rebalanced mix; retain a mute only when removal is the intended output.

Practical starting moves for a rebalanced mix

These are audition ranges, not recommended final settings. Genre, mastering, stereo effects and arrangement determine how far a source can move before leakage becomes obvious. Keep playback loudness comparable when judging the result; a louder version can seem better merely because it is louder. Keep the smallest adjustment that solves the problem.

Create and work with four stems

1. Return to the unprocessed source. Select the whole song for full-length stems, or select the exact region you need: Stem Split processes the selected audio.
2. In Music Rebalance, set Sensitivity and Quality, then click Stem Split. Alternatively, open the top-right Stem Split pane, choose its Music Rebalance minimodule, set those controls and click Process.
3. In the resulting Stems tab, choose All Stems from the Stem selector at the left of the tab header to see the four lanes.
4. Audit each lane before repairing it. Click an already active lane selector to leave only that lane active. If the target lane is inactive, double-click its selector to activate only that lane. Double-click an active selector to activate all lanes again.
5. Before selecting a region or applying another module, check which lanes are active: playback, selections and edits affect only those lanes. A repair intended for Voice should not also process Drums.

The Stems View controls and export instructions also explain two easily missed differences. Playback in All Stems is attenuated by Preferences → Audio → Composite view gain reduction. Set that value to 0 dB to hear the clip at the original tab’s level. This is a monitoring setting; do not normalize each stem to compensate for quieter playback.

Module Compare buttons are unavailable in All Stems. To compare a repair on Voice, select Voice using the tab header’s Stem selector, rather than merely leaving its lane active in All Stems. You can then edit that stem like an ordinary file. Edits remain when you return to All Stems.

For export, choose the command that matches the tab view:

• All Stems → File → Export…: export all stems as separate files.
• All Stems → File → Export as Single File…: export their sum as one file.
• One stem selected → File → Export…: export that stem alone. Use File → Export Selection… for a selected region of it.

For a DAW handoff, use the same full-song boundaries and sample rate for every stem. Keep leading silence and avoid trimming each file independently. Check each stem’s sample and true peaks before integer export, using the Waveform Statistics and Gain procedure below. If an aligned stem set needs attenuation, reduce every stem by the same amount to preserve its balance; that gain-adjusted sum will no longer null against the original. Reopen the exported files and check peaks, alignment, length and the beginning and ending before delivery. Label them as derived Voice, Bass, Drums and Other stems, rather than original multitracks.

Editor, real-time plug-in or ARA

Music Rebalance is in RX 12 Standard and Advanced, not Elements. The official RX 12 edition matrix confirms the boundary. Use the editor for the Compare and Stems View procedures in this guide, separate export and follow-up spectral repair. A real-time plug-in is useful for asking whether a small balance move helps before you print it.

The current manual lists VST3 and AU ARA, plus AudioSuite and AAX real-time availability in Pro Tools. Host and format determine the available workflow; do not assume the editor’s Stems tab or Compare window will appear inside every plug-in. If the plug-in is missing, use the RX plug-ins not showing checklist to confirm edition, format and scan state.

Limits to hear before delivery

Dense distortion, layered vocals, shared reverbs and instruments with similar spectra make clean boundaries difficult. Vocal reverb may remain in Other after dry Voice is attenuated. Kick can travel with Bass, and bright instruments can move with Drums. Boosting a separated stem also boosts its leakage.

Check the opening, busiest chorus, breakdown and ending. Algorithms can behave differently as the arrangement changes. If the instrumental sounds acceptable in the verse but collapses in the chorus, either process sections with different settings or reduce the ambition of the move.

iZotope’s RX 12 stem-separation demonstration shows the unchanged stems recombining to null against the original. That demonstrates reconstruction of the mix, not purity of each stem or recovery of the recording session. Leakage can cancel when stems are summed yet become audible when you change their balance. Once you edit or relevel them, do not expect the same null result.

RX producing files does not itself grant permission to distribute someone else’s stems or remix. For US copyright questions, the Copyright Office’s guidance on permission and derivative works explains why permission or an applicable exception is a separate issue.

A complete Music Rebalance workflow

For one rebalanced mix, use this sequence in the RX editor after the listening checks above:

1. Keep the difficult selection unprocessed. With the same stem Soloed, use Compare to add separate candidates for the settings you want to audition. Name each entry, select it in Compare Settings and Preview it. Repeat with another stem when its separation matters.
2. Select the preferred candidate and click View settings to load its settings into the module. Clear the audit Solo/Mute controls and set the intended gains. For an instrumental, retain only the intended removal.
3. Click Compare again to create a new candidate containing those final full-mix settings. Preview that entry against the original. An earlier Soloed candidate does not automatically update when you change the module.
4. If you want to apply the short test, select this final entry and click Render in Compare Settings. Listen in context, then use Undo History to return the file to its pre-test state before processing the whole song.
5. Load the chosen final settings back into the module with View settings. Select the full song and Render once from the module. Check the opening, chorus, breakdown and tail; undo and revise if a setting that helped the test harms another section.
6. After rendering, select the full file and check Sample Peak Level and True Peak Level in Waveform Statistics. A stem boost can push a finished master over full scale. Before integer WAV/AIFF export, leave headroom below 0 dBFS for sample peaks and below 0 dBTP for true peaks. If necessary, use the Gain module with a negative dB adjustment on the whole file, including both stereo channels and the full frequency range; Render that reduction and measure again. This lowers the overall result without changing its internal balance.
7. Use File → Export… to write a new WAV or AIFF with a distinct name. Confirm the export options, then choose the destination and save. Reopen the result and check the actual delivered file. File → Save RX Document… preserves the editable RX session and history; ordinary Save can overwrite an opened WAV or AIFF.

The RX saving and export guide distinguishes an RX document from an audio file for other applications. If you need four files instead, return to the original and follow the Stem Split procedure above.

For edition choices, the RX 12 review covers the wider product. For another music repair after separation, use the Music and Vocals hub; the RX Workflows hub covers broader processing sequences.

Frequently asked questions

Which stems does RX 12 Music Rebalance create?

It separates Voice, Bass, Drums and Other. Other contains content not classified in the first three groups, typically melodic instruments.

Can Music Rebalance make a clean instrumental?

It can make a useful instrumental, but vocal reverb and overlapping centered instruments may remain or leave holes. Test the busiest chorus before processing the full song.

Does Music Rebalance separate guitar and piano?

Not as dedicated named stems. Guitar, piano, strings and other melodic material are generally grouped into Other unless parts are classified with another category.

What is the difference between Render and Stem Split?

Render applies the gain, mute and solo decisions as one resulting mix. Stem Split creates four independent stems and ignores the module’s gain slider positions, using 0 dB for the split.

How should I set Music Rebalance Sensitivity?

Increase only the stem that is missing useful detail, then listen for added leakage. Sensitivities are relative, so raising every stem to 100% does not improve the separation.

Which Music Rebalance quality mode should I use?

Use Good/Real-time for a fast feasibility test, then compare Better/Offline and Best/Offline on the same Soloed stem and selection. Best is slowest and includes anti-phasing processing. Clear audition Solo and Mute controls before rendering a normal rebalanced mix.

Is Music Rebalance included in RX Elements?

No. The current RX 12 documentation lists Music Rebalance for Standard and Advanced.

Are RX Music Rebalance stems the same as original multitracks?

No. They are machine-learned estimates derived from a mixed file. They can be useful for repair and rebalance, but leakage and missing detail may remain.