How to Remove Vocals with iZotope RX 12

To remove vocals with iZotope RX 12, use Music Rebalance in the Audio Editor supplied with Standard or Advanced. Work on a copy and turn Instant Process off before selecting a difficult chorus. Choose Best/Offline, mute Voice and leave Bass, Drums and Other unmuted at 0 dB. Compare vocal leakage with lost accompaniment, then process the untouched full song once and export a new instrumental file.

Before committing, unmute Voice and solo it to hear what RX would remove. Turn Solo off and mute Voice again to hear the instrumental. If guitars or cymbals disappear with the singer, the setting is too costly for that passage. Judge the backing track by how naturally it plays, including the spaces between vocal phrases.

What RX can and cannot remove from a finished mix

Music Rebalance analyses a finished mix and classifies its content as Voice, Bass, Drums and Other. The current RX 12 Music Rebalance manual says each element can then be adjusted independently. Muting Voice creates the instrumental sum of the remaining three classifications.

This is source separation, not access to the original multitrack. Vocals share frequencies, transients, reverb and stereo space with instruments. RX can often produce a useful backing track, remix reference or rehearsal version, but it cannot promise the untouched instrumental master that existed before mixing.

For changing the balance of several instruments or exporting every stem, follow the RX 12 Music Rebalance guide. Here the useful comparison is narrower: how much vocal can you remove before the accompaniment suffers?

Start with the best source you legally have

Use the highest-quality authorized stereo source available, ideally lossless. MP3 or streaming compression does not make separation impossible, but it gives the model codec residue to classify along with the music. Converting an MP3 to WAV does not restore discarded detail; it only puts the same decoded audio in a lossless container.

Preserve the original and work on a copy. Keep the source sample rate unless the destination requires another rate. Start without extra normalization or limiting so you can compare against the supplied mix without another processing variable. Those processes will not restore the original multitrack.

Before drawing a selection, turn Instant Process off: when enabled, it can apply another module as soon as you select audio. Then choose a chorus containing lead, harmonies, reverb, dense instruments and cymbals. Include the transition into or out of it so you can hear changes when the singer begins.

Use Render or Stem Split for the instrumental?

Use the full Music Rebalance module and Render when you only need one no-vocal mix. The RX 12 Stem Split manual explicitly recommends the module route for muting, soloing or changing element levels. Mute Voice, audition the remaining mix and render that result.

Use Stem Split when you want separate Voice, Bass, Drums and Other files for later editing or mixing. Stem Split opens them in a Stems tab, where each can be processed and exported independently. It is useful when the instrumental needs targeted cleanup after separation.

Do not move the module's gain sliders and expect Stem Split to print those moves. The manual says splitting ignores individual gain positions and assumes 0 dB. Its output still includes a Voice stem. Use Render for the one-file instrumental route, or deliberately leave the exported Voice file out of your later mix; do not treat a module mute as proof that the split output is vocal-free.

Nine steps to remove vocals in RX 12

1. Open a working copy. Open the highest-quality authorized stereo source in RX 12 Standard or Advanced Audio Editor. Preserve the original; turn Instant Process off before making selections.
2. Select the hardest section. Select a chorus and its transition, including backing vocals and reverb tails. Keep this test short enough to compare settings on the same original audio.
3. Open Music Rebalance. Use the full Music Rebalance module for a single instrumental mix, rather than starting with the Stem Split pane.
4. Choose Best/Offline. Select Best/Offline when processing time is acceptable. Compare Better/Offline if the musical result or turnaround requires a different compromise.
5. Set the instrumental balance. Mute Voice. Leave Bass, Drums and Other unmuted at 0 dB and clear any active Solo buttons.
6. Tune Voice Sensitivity. Change Voice Sensitivity relative to the other stems only as needed. More vocal capture can also remove more instruments; use the same untouched test audio for each comparison.
7. Check both listening states. To check what will be removed, unmute Voice and solo it. Then turn Solo off and mute Voice again for the instrumental. Audition through Preview or Compare where available, or render a disposable test copy; ordinary playback of an unrendered original is not the processed comparison.
8. Process the full song once. Return to an untouched full-song working copy, or undo every test render in the working file. Confirm the accepted quality and Sensitivity, clear Solo, mute only Voice, select the entire song and Render once. Check the intro, verses, choruses, bridge and ending.
9. Export the instrumental. Use File > Export to save a new WAV or AIFF at the required sample rate and bit depth. Check peaks and preserve headroom; verify the full exported file in a player or DAW outside RX.

A rendered test changes the working audio. Do not repeatedly split that changed section while choosing settings, and do not include it in a second whole-song render. Use untouched copies or Undo between trials. The Preview and Compare controls can help audition alternatives without committing each one; Compare's Render command does apply its selected result.

Choose Best/Offline for the final pass

RX 12 offers Good/Real-time, Better/Offline and Best/Offline. Good is the fast mode for real-time work. Better improves separation with a modest speed cost. Best is the slowest and includes anti-phasing processing. For a final instrumental, start the serious comparison with Best unless turnaround time or hardware makes it impractical.

Use Good to confirm routing or rehearse a level move. For a difficult passage, compare the same original 20–30 seconds in Better and Best; that duration is a suggested test size, not a processing requirement. Start each rendered comparison from an untouched copy. The highest documented separation quality may still be a less useful musical compromise on a particular mix.

The AudioSuite version in Pro Tools cannot preview Music Rebalance before rendering, according to the current manual. For careful vocal removal, a short duplicate clip or the standalone Editor makes comparison easier. The RX Editor and Connect workflow explains how to move selected material safely between environments.

Set Voice Sensitivity without stealing instruments

Sensitivity tells the separation algorithm how much attention to give each stem. Raising Voice Sensitivity can capture quieter breaths, consonants, harmonies and reverb that otherwise remain in the instrumental. It can also classify more instrument energy as Voice, which disappears when Voice is muted.

Treat the four Sensitivity controls as a balance. They operate relative to one another; setting all four to 100% does not make separation more intense. Change Voice against Bass, Drums and Other, then audition the same original test rather than processing an already separated result again.

Use two listening states. First, unmute Voice and solo it. Listen through the module's available audition controls, or render a separate test copy; note any recognizable cymbal wash, guitar chords or synth pads. Next, clear Solo and mute Voice for the instrumental comparison. Confirm Bass, Drums and Other are unmuted at 0 dB. Move Sensitivity toward the least objectionable compromise rather than chasing complete silence.

Why lead vocals or backing vocals remain

The vocal tail is mixed like an instrument. Reverb and delay may spread wide and overlap pads, guitars or cymbals. A setting that captures every tail can remove too much of those instruments. A faint tail may be the more natural compromise.

Backing vocals occupy a different position. Wide harmonies, gang vocals and ad-libs do not always classify with the centered lead. Test sections where they sing without the lead; adjust Voice Sensitivity there and compare the cost to the mix.

The master is dense or heavily processed. Source separation estimates components; it does not reverse every compression, saturation or limiting decision in a finished mix. iZotope's 2025 RX 11 Music Rebalance examples demonstrate practical no-vocal uses. They are historical examples, not RX 12 benchmarks or a guarantee for your master.

You tested a quiet verse only. Settings can pass on one singer over piano and fail on the layered chorus. Keep the settings only if that difficult section sounds usable, then spot-check each section of the full render.

Why the instrumental sounds hollow, phasey or watery

A hollow midrange can be a sign that instrumental content was classified as Voice and removed with it. Check the soloed Voice against the original to test that explanation. Try lower Voice Sensitivity or more relative attention to the competing stem, then compare again. Broad EQ cannot reconstruct an instrument that separation has removed.

For swirling or watery highs, compare with the original first. If the artifact appears only after separation, try Best/Offline and a less aggressive Voice classification. If it is already present in the source, try a lossless original if available. The sound alone does not prove whether cymbals, reverb, codec artifacts or separation caused it.

Pumping at phrase boundaries can appear when accompaniment shares the vocal envelope. Listen from one bar before the phrase through one bar after it. A static mute may reveal the change more clearly than listening only inside the lyric.

The RX spectrogram tutorial helps locate a leftover phrase or damaged frequency region. Avoid painting out every visible harmonic: once the separated instrumental sounds natural, visual residue is not automatically a defect.

Clean difficult residuals without overprocessing

If one ad-lib or breath remains, preserve the rendered instrumental and try a narrow correction on a copy. Spectral Repair may help a localized residual, but a selection containing guitar or cymbal energy can damage those sounds too. Undo it if the accompaniment loses more than the vocal residue justifies.

If you need access to the separate components, split an untouched copy into stems and select a particular stem in the Stems tab for editing. The manual allows that stem to be processed like another audio file. Repeated vocal reverb is a reason to compare Sensitivity again on the original, not automatically stack denoisers on the instrumental. For a few exposed syllables, small manual attenuation may be a better compromise.

Keep a version before cleanup and compare at matched level. Extra processing can make the instrumental quieter or smoother when heard alone, yet leave it sounding thin when a new singer is added. Test in the actual use: rehearsal room, remix session or edit bed.

Export the instrumental correctly

After the module Render route, use File > Export to write a new file, normally WAV or AIFF. Match the source sample rate unless the destination requires another rate, and choose a bit depth appropriate to delivery. Check peaks rather than normalizing blindly: removing one component can change the summed peak level. Save an RX Document separately if you need the editing history; an RX Document is not the audio file to deliver to a player or DAW.

With Stem Split, the stem selector in the tab header chooses an individual component; File > Export then exports that stem. From All Stems view, Export writes separate files. For a backing track, bring the exported Bass, Drums and Other files into your DAW at the same start position and leave Voice out. RX also has Export as Single File, documented as summing all stems. Do not assume that a lane deactivated for listening is excluded from that export; use the module route or an explicit three-stem DAW mix for this tutorial.

Name the result clearly with source title, “instrumental,” RX version, date and sample rate. Reimport the file into a blank session and play the beginning, loudest chorus and ending. That catches wrong stem selection, truncation, channel mistakes and export settings that were invisible inside the working session.

Can you keep backing vocals but remove the lead?

Music Rebalance has one Voice category, not separate lead and backing-vocal faders. If the backing vocals classify as Voice, muting Voice removes them too. If they leak into Other, some may remain—but not as a controllable promise.

A practical attempt is to split stems, keep the full Voice stem as a reference, then selectively restore wanted harmony regions into the instrumental where they do not overlap the lead. That becomes manual spectral and arrangement editing, not a one-click vocal remover. Ask for original stems when the distinction matters commercially.

Likewise, removing one speaker while keeping another is not a separate target offered by this four-way music model. For other tasks, use the Music & Vocals guides to find vocal extraction, vocal cleanup and music-separation workflows.

Copyright and permission still apply

Software access is not music clearance. The U.S. Copyright Office's music copyright guide distinguishes the composition from the sound recording; they can have different owners and rights. Copying, distribution and derivative uses can require permission, an applicable statutory licence or an exception. The guide distinguishes the composition's public-performance right from the recording's more limited digital-audio-transmission right.

Do not assume that personal or noncommercial use automatically qualifies as fair use. The Copyright Office's Fair Use Index explains that the circumstances must be considered together. Rules also differ by country. Check the rights for your actual karaoke release, remix, performance or client video, and obtain legal advice where needed.

A practical quality gate for the finished instrumental

• No intelligible lead words remain in exposed sections.
• Breaths, reverb and delay tails are not more distracting than the original vocal.
• Snare, cymbals, guitars, keys and synth pads retain believable attacks and tone.
• Bass weight and stereo image remain stable across vocal entries.
• The full song has been checked, not only the test chorus.
• The exported file opens correctly, has the intended channels and preserves headroom.
• The planned use is authorized.

If words or tails remain, test slightly more Voice capture against the cost to the accompaniment. If attacks, tone or stereo stability suffer, back off and compare again. Sometimes a faint residual is the better result; otherwise seek an authorized instrumental or original stems. Music Rebalance requires Standard or Advanced for this Editor workflow; see the RX edition comparison if you have Elements.

Frequently asked questions

Can iZotope RX completely remove vocals?

It can create a useful instrumental from many mixes, but complete removal without instrument damage is not guaranteed. Reverb, backing vocals and overlapping instruments often force a compromise between residue and missing accompaniment.

Which RX module removes vocals?

Use Music Rebalance in RX 12 Standard or Advanced Audio Editor. Mute Voice while keeping Bass, Drums and Other unmuted. Render applies that balance; File > Export creates the delivered instrumental. Stem Split is useful when you need separate components for editing.

What Music Rebalance quality should I use?

Use Best/Offline for the final comparison when render time is acceptable. Good/Real-time is fastest, Better/Offline is a middle option, and Best includes anti-phasing processing but is slowest.

Should I turn Voice Sensitivity to 100%?

Not automatically. Higher Voice Sensitivity can capture more breaths and vocal tails but also pull instruments into the muted Voice stem. The controls are relative, so compare the soloed Voice and the remaining instrumental.

Why can I still hear the singer after muting Voice?

Vocal reverb, delay, harmonies or phrases may remain in the other classifications. Compare slightly higher Voice Sensitivity on the original, or try a narrow correction on a copy of the instrumental; stop if the accompaniment becomes hollow.

Can RX remove lead vocals but keep backing vocals?

Not with a dedicated one-click control. Music Rebalance has one Voice category. A manual attempt can restore wanted harmony regions from a separated Voice stem where they do not overlap the lead; it is not a guaranteed separation of two singers.

Should I export stems or one instrumental file?

Use Music Rebalance Render followed by File > Export for one instrumental. Export separate stems when you plan more editing or mixing, then deliberately omit Voice from the backing-track mix. Stem Split ignores individual module gain positions and separates at 0 dB; its output includes Voice.

Can I publish a karaoke track made from any song?

Not automatically. Songs and recordings are protected works, and public distribution, performance or derivative use may require permission or licences. Check the rules for your country and planned use.