> ## Content Index > Fetch the complete content index at: https://izotoperx.net/llms.txt > Use this file to discover other available public pages before exploring further. # iZotope RX 12 Scene Rebalance: DME Separation Guide - URL: https://izotoperx.net/izotope-rx-scene-rebalance/ - Published: 2026-09-04T09:52:28.000Z - Updated: 2026-09-08T07:59:09.000Z - Description: A practical Scene Rebalance workflow for rescuing dialogue, music and effects from a flattened mix without mistaking machine-learned separation for original stems. - Author: Brandon Hayes - Tags: Music Repair, RX 12, Post Production **Use iZotope RX 12 Scene Rebalance when you have a finished stereo soundtrack but need to change the balance of dialogue, music and effects.** Start from the default separation, solo each stem only to find leakage, adjust the relative Sensitivity controls before making large Gain moves, and approve the result with all three components playing together. The steps below use the standalone RX Audio Editor. Split to Stems View when one component needs local repair, replacement or separate export. Scene Rebalance is an RX 12 Advanced feature. It is rescue technology, not a substitute for original production DME stems. A two-decibel dialogue lift that saves a scene can be a better result than trying to remove the score completely and exposing separation artifacts. ## Scene Rebalance at a glance | Job | Best route | Main risk | | ------------------------------------ | ---------------------------------------------------------------------------------- | ------------------------------------------ | | Lift buried dialogue under music | Voice Gain plus sensitivity check | Music/ambience leaking into Voice | | Reduce a baked-in score | Music Gain, modest attenuation | Dialogue texture and tonal holes | | Reduce overly loud impacts | Effects Gain or local Effects-stem repair | Dialogue transients classified as Effects | | Repair lav rustle only in speech | Stem Split, then [De-rustle on Voice](https://izotoperx.net/izotope-rx-de-rustle/) | Separation leakage processed with the stem | | Replace licensed music | Stem Split, edit Music, recombine | Residual music in Voice/Effects | | Clean noise and reverb around speech | [Dialogue Isolate](https://izotoperx.net/izotope-rx-dialogue-isolate/) | Using the wrong three-way split | | Separate vocals, bass, drums, other | [Music Rebalance](https://izotoperx.net/izotope-rx-music-rebalance/) | Confusing music stems with DME | ## What Scene Rebalance actually separates The [current RX 12 manual](https://docs.izotope.com/rx12/en/scene-rebalance.html?ref=izotoperx.net) says Scene Rebalance uses machine learning to identify Voice, Music and Effects in a soundtrack. Each component has Gain, Mute, Solo and Sensitivity controls. This is a program-mix problem: the original production tracks or stems are unavailable, but the balance needs intervention. “Voice” is the dialogue-like content the model assigns to speech. “Music” is score or musical material. Treat Foley, impacts and ambience as material to inspect across the split; do not assume every such sound will land in Effects. The categories are estimates, not metadata recovered from the old session. A door slam can share transients with music; reverberant dialogue can leak into other components. The [official RX 12 feature overview](https://www.izotope.com/community/blog/rx-12?ref=izotoperx.net) classifies Scene Rebalance as new and Advanced-only. The plug-in format is AudioSuite only, according to current release notes, so plan offline clip processing or the standalone editor rather than a universal real-time insert. ## When Scene Rebalance is the right first tool Use it when a client sends a stereo master instead of DME stems, an archive contains only the final broadcast mix, or a social-video export has dialogue buried under score. It can also create a working music component for replacement when licensed music must change and the source session is gone. Do not use it merely because the file contains dialogue. Stable background noise belongs in a denoise workflow; speech plus changing noise/reverb belongs in Dialogue Isolate; music vocals/bass/drums belong in Music Rebalance. If original production stems exist, request them. They preserve the mixer’s source boundaries and usually beat model-derived separation. Our [RX 12 new-features guide](https://izotoperx.net/izotope-rx-12-new-features/) explains how Scene Rebalance fits the release. Use the steps here to choose settings and check the resulting soundtrack. ## Protect the flattened source and delivery facts Duplicate the source and preserve channel layout, sample rate, timecode, sync start and duration. Use [File > Save RX Document](https://docs.izotope.com/rx12/en/working-with-files.html?ref=izotoperx.net) for the RX working session and its edit history. Ordinary Save on a WAV or AIFF overwrites that file; export a separately named deliverable. Write down the requested change: “dialogue +2 dB in scene 14” is testable; “make it cleaner” is not. Record whether the client expects a new stereo mix, separate derived stems, or material for a later remix. Never label machine-separated files as original D, M and E production stems. Use names such as `derived-voice`, `derived-music` and `derived-effects`, with the RX version and date in the session notes. That prevents a downstream editor from confusing reconstructed components with the original tracks. An unchanged split may sum back to the input while its individual stems still contain cross-classified sounds. ## Choose a representative test scene Select a scene containing the hard overlap: a quiet line under score, an effect crossing a word, ambience, music transients and at least one clean phrase. A clean isolated sentence proves very little. Include the point where the mix enters and leaves the problem so gain or artifact changes cannot hide at the selection boundary. If the program changes style—interview, action sequence, montage, exterior ambience—build separate test regions. One sensitivity balance will not necessarily fit all scenes. Use consistent pre-roll and save an RX Document before committing another repair; our [Preview and Compare workflow](https://izotoperx.net/izotope-rx-selections-preview-compare-history/) covers the discipline. ## Preview the default split before changing controls Listen first with all stems at 0 dB. Then solo Voice, Music and Effects one at a time. Write down missing wanted content and intrusive leakage for each. A breath missing from Voice may be fixable with Voice Sensitivity. A snare leaking into Voice may require relatively more Music/Effects attention or a smaller final gain move. Do not approve from solos. Separation often sounds rough when a component is isolated but recombines acceptably because complementary material returns from the other stems. Conversely, editing or attenuating a component can remove content needed by the full mix even if its solo seems clean. ## Sensitivity controls are relative, not three independent quality knobs Each Sensitivity control changes how strongly the algorithm attends to that stem. Raising Voice can retain subtler speech details such as breaths, but it can also bring music or effects leakage into Voice. The manual makes a counterintuitive rule explicit: setting all three sensitivities to 100 percent is not a quality upgrade; the values operate relative to one another. Change one relationship at a time. If quiet speech is missing, raise Voice slightly or lower competing attention, then replay the same line. If score transients appear in Voice, reverse the last move and test Music sensitivity. Keep a note of the three values; otherwise several small changes become impossible to audit. Sensitivity determines classification. Gain determines the level after separation. Do not use Gain to conceal a classification error until you have tested whether Sensitivity can improve it. ## Make the smallest gain move that solves the scene Voice, Music and Effects Gain operate in decibels. Start with the editorial request, not the maximum range. For example, try a one- or two-decibel Voice lift or a modest Music cut, then listen at delivery level; these are audition examples, not a prescribed setting. Large moves reveal leakage and missing spectral content that a rescue-sized move leaves masked. Check transition points where dialogue starts and music swells. A static gain may solve the center of the scene but feel wrong around the edges. Render a scene with sufficient surrounding context or automate the exported stems downstream when the requested balance changes through time. Do not force one global setting across a reel. ## Use Mute and Solo as diagnostic tools Solo answers “where did this sound go?” Mute answers “which component is carrying the objectionable material?” Use them to find cross-classification: dialogue consonants in Effects, score pads in Voice, Foley impacts in Music. Keep the full mix as the approval reference. When replacing music, mute Music and listen for score residue in Voice and Effects. If the residue is obvious, a clean replacement may mask some of it, but do not promise a music-free DME. The official [Scene Rebalance walkthrough](https://www.izotope.com/community/blog/scene-rebalance?ref=izotoperx.net) demonstrates the replacement use case; success remains source-dependent. ## Choose direct Rebalance or Stem Split Use the direct module when the goal is a simple level rescue: dialogue up, score down, effects restrained. Use Stem Split when a component needs local spectral repair, different processing across time, replacement or separate delivery. More editing flexibility also means more chances to damage sync or recombination. The RX 12 [Stem Split pane manual](https://docs.izotope.com/rx12/en/stem-split-pane.html?ref=izotoperx.net) describes two entry points: use Stem Split in the Scene Rebalance module footer, or open Stem Split at the editor’s top right, choose Scene Rebalance, adjust Sensitivity and click Process. Both split the selected audio. Select the whole file when you need full-length stems. Our [Stems View tutorial](https://izotoperx.net/izotope-rx-stems-view/) covers track selection, processing and export beyond this module. ## Stem Split ignores the visible Gain sliders This is the easiest mistake to make. The Scene Rebalance manual states that Stem Split creates Voice, Music and Effects at 0 dB and ignores the module’s individual Gain-slider positions. If you previewed Music at −6 dB and then split, the new Music stem is not automatically −6 dB. Record the approved gain relationship before splitting, then apply it with the [Gain module](https://docs.izotope.com/rx12/en/gain.html?ref=izotoperx.net) to the intended stem, or use faders in the receiving DAW. All Stems has lane selectors, not a set of mixer faders. Check the unchanged three-stem sum against the original before adding new level decisions. In Stems View, Preferences > Audio > Composite view gain reduction attenuates monitoring; set it to 0 dB for an equal-level baseline, with a safe listening volume. ## Repair only the affected stem Choose Voice in the tab’s Stem selector for a lav scrape, mouth click or isolated dialogue problem. In All Stems, selections, playback and edits affect the active lanes. A click on an active lane isolates it; double-click an active lane to reactivate all stems. Confirm lane highlights before processing and before judging the sum. Apply the lightest module or Spectral Repair selection that solves it. Do not denoise Music and Effects just because they are visible. A defect created by separation can worsen when another ML processor tries to “clean” it. For an isolated event, the [Spectral Repair guide](https://izotoperx.net/izotope-rx-spectral-repair/) covers attenuation and reconstruction. For dialogue-wide noise, follow the [Clean Dialogue hub](https://izotoperx.net/clean-dialogue-rx/). Return to the full three-stem playback after every material edit. ## Recognize leakage and separation artifacts - **Voice texture:** watery consonants, missing breaths, hollow vowels or ambience glued to words. - **Music damage:** smeared attacks, chorused pads, vocal ghosts or unstable stereo width. - **Effects damage:** dialogue transients, holes around impacts or room tone that pumps. - **Recombination damage:** phasey center, transient duplication, tonal gaps or changed loudness. When artifacts appear, reduce the requested gain move, revisit relative Sensitivity or patch from an alternate source. Avoid reducing the processing selection to a tiny click merely to make it faster; test a larger scene with surrounding material, then check the rendered result before keeping it. “Residual music but natural speech” may be the professional compromise. A [Scene Rebalance user report about Preview and Render disagreeing](https://www.reddit.com/r/iZotopeAudio/comments/1ut6mjd/scene%5Frebalance%5Fyields%5Fdifferent%5Fresult%5Fthan%5Fwhat/?ref=izotoperx.net) describes better agreement with a longer selection. That is one unverified report, not a confirmed cause or a guaranteed fix. It is a reason to audition the committed audio as well as Preview. ## Recombine before approving any stem Play Voice, Music and Effects together at the intended gains. Match loudness against the original. Check dialogue intelligibility, score continuity, impact, ambience and stereo image. A clean Voice solo does not matter if the full mix becomes smaller or phasey. During module Preview, compare with Bypass from the same pre-roll. After Render or stem edits, compare with the untouched source or an earlier History state; Bypass does not undo committed processing. Compare on full-range monitors and a small speaker because dialogue masking may reappear at lower playback levels. Check headphones for spectral swirls that speakers and room acoustics can hide. ## Export and label the result safely Export to a new location and preserve original duration, channel layout, sample rate and sync reference unless the delivery specification says otherwise. Use lossless working files, such as WAV. In All Stems, File > Export writes separate stems; File > Export as Single File writes their sum. To export just one component, select that stem in the tab header first. Export Selection writes only the chosen range. Reopen the outputs and check their content and start alignment; do not infer the delivery from which lane was last soloed. If derived stems are delivered, identify the separation tool/version and whether level moves are baked in. Scene Rebalance changes program energy, so remeasure loudness and true peak after recombination. A dialogue lift can raise integrated loudness and peaks even when the scene sounds subjectively balanced. Verify the final encoded or wrapped deliverable, not only the editor timeline. For handoffs to common video editors, the [DaVinci Resolve/Fairlight workflow](https://izotoperx.net/izotope-rx-davinci-resolve-fairlight/) and [Premiere Pro RX workflow](https://izotoperx.net/izotope-rx-premiere-pro/) cover round trips and render checks. ## Choose the AudioSuite or standalone route deliberately The current [RX 12 release notes](https://www.izotope.com/pages/release-notes/rx-standard?ref=izotoperx.net) list Scene Rebalance as AudioSuite-only in plug-in form. The Scene Rebalance AudioSuite plug-in cannot preview processing before rendering. In Pro Tools, preserve the unprocessed clip on a duplicate playlist or in a separate source copy, render a test selection with sufficient context, and audition that result against the source. Check the AudioSuite output options and the processed clip’s placement before keeping it. An AudioSuite render commits the chosen relationship. The standalone editor is better when you need the full spectrogram, several Compare candidates or Stem Split followed by local repair. Transfer the exact time range, preserve timecode/reference sync, and return a versioned render. Choose a bounded scene with enough surrounding audio to test transitions; the best selection length depends on the material. Whichever route you choose, follow the wider signal-flow logic in iZotope's [audio post-production workflow](https://www.izotope.com/en/learn/audio-post-production-workflow-101?ref=izotoperx.net): repair serves the edit and final mix, not the other way around. Keep picture, scene intent and delivery requirements visible while judging the separation. ## Use Preview, Compare and History as evidence In the standalone editor’s normal file tab, the [Preview and Compare controls](https://docs.izotope.com/rx12/en/common-module-controls.html?ref=izotoperx.net) let you audition Scene Rebalance settings before committing them. Compare a small Music cut at default Sensitivity with the same cut at revised Voice Sensitivity. Treat a later Stem Split and local repair as a separate saved candidate. Compare and Clip Gain are unavailable in All Stems; use the tab’s Stem selector to open an individual stem when those features are needed. Give every candidate the same pre-roll and loudness. Listen through dialogue entries, effects impacts and score transitions rather than looping a single vowel. After a render, double-click an item in [Undo History](https://docs.izotope.com/rx12/en/undo-history.html?ref=izotoperx.net) to name it with the three sensitivity values, gain moves and region. Save an RX Document to retain that history. Preview alone is not a saved edit. If a later edit sounds worse, you should be able to return to the exact approved candidate rather than rebuilding it from memory. ## Ten-step Scene Rebalance workflow 1. **Protect and identify the source.** Keep the stereo mix untouched, confirm real stems are unavailable and save a separate RX Document. 2. **Select a representative scene.** Include overlap, quiet dialogue, music detail, effects and ambience. 3. **Preview the default separation.** Listen to the mix, then solo each component to map leakage. 4. **Adjust relative Sensitivity.** Change one stem’s attention relative to the others, checking wanted detail and unwanted leakage. 5. **Set modest Gain changes.** Use the smallest D/M/E move that solves the scene. 6. **Compare and choose the route.** Decide between direct rebalance and Stem Split at matched loudness. 7. **Split to Stems View when local repair is needed.** Split the intended time range, remember each stem starts at 0 dB, and verify active lanes before repair. 8. **Recombine and check artifacts.** Judge leakage, texture, transients and ambience with all stems together. 9. **Export and label honestly.** Preserve the source and identify outputs as derived stems. 10. **Verify the delivery.** Reopen the rendered output and check sync, layout, duration, loudness and true peaks. ## Final quality-control checklist - Test a difficult overlap, not only a clean dialogue line. - Write down all three Sensitivity and Gain values. - Check each solo for leakage, then approve only the recombined mix. - Confirm Stem Split ignored Gain and starts at 0 dB. - Inspect speech, music transients, effects and ambience. - Compare at equal loudness on speakers and headphones. - Preserve sync, duration, sample rate and channel layout. - Label derived stems and retain the original mix. The [Music & Vocals in RX hub](https://izotoperx.net/music-vocals-rx/) connects Scene Rebalance with music stem repair, De-bleed, vocal cleanup and restoration workflows. ## What not to do - Do not raise all three sensitivities and expect better separation. - Do not confuse Sensitivity with Gain. - Do not approve a scene from soloed stems. - Do not assume Stem Split carries the visible Gain settings. - Do not label separated output as original DME stems. - Do not chase complete music removal when speech becomes synthetic. - Do not overwrite the received mix. - Do not skip loudness, true-peak and sync checks after rebalance. ## Frequently asked questions ### What is iZotope RX Scene Rebalance? It is an RX 12 Advanced module that uses machine learning to separate Voice, Music and Effects from a soundtrack so their levels can be changed independently. ### Is Scene Rebalance included in RX 12 Standard? No. Scene Rebalance is Advanced-only. Its plug-in format is AudioSuite, without Preview before Render. The standalone Audio Editor also provides the module and Stem Split route. ### How do Scene Rebalance Sensitivity controls work? They change the algorithm's relative attention to Voice, Music and Effects. Raising one can retain subtler content but also increase leakage. Setting all three to 100 percent is not a quality upgrade; their relative balance matters. ### What is the difference between Sensitivity and Gain? Sensitivity changes which material the model classifies into a stem. Gain changes that separated stem's level in decibels. Fix classification before using extreme level moves. ### Does Stem Split use my Scene Rebalance Gain settings? No. The RX 12 manual says Stem Split ignores the three Gain positions and creates Voice, Music and Effects at 0 dB. Recreate the approved balance after splitting. ### Can Scene Rebalance remove music completely? It can reduce or isolate music, but complete transparent removal is not guaranteed. Music may leak into Voice or Effects, and extreme attenuation can expose speech artifacts. ### How is Scene Rebalance different from Music Rebalance? Scene Rebalance separates Voice, Music and Effects from a soundtrack. Music Rebalance separates vocals, bass, percussion and other instruments from music. ### Are Scene Rebalance outputs real DME stems? They are machine-derived dialogue, music and effects components, not the original production stems. Label them as separated outputs and request original DME whenever it is available.