Audio Repair and Restoration
How to remove background music from a recording
Published September 3, 2026
You cannot cleanly wipe music out of a mixed voice recording with normal noise reduction, because music is not steady background noise. Pulling the voice out from under music is a separation job, and the realistic tool for it is AI stem separation. Run the file through a free separator such as Ultimate Vocal Remover, take the isolated vocal it gives you, then clean that up in REAPER. Expect a big improvement on clearly produced music and a much rougher result when the voice and music sit on top of each other. If you have the original session or the music on its own track, use that instead, because it will always beat separating a finished mix.
People ask this in two very different situations, and the fix depends on which one you are in. Either the music and the voice are on separate tracks in your project and you just want to mute or lower the music, or the music and the voice are baked together into one finished file and you want to lift the voice back out. The first case is easy. The second case is the hard one, and it is the one this guide is really about.
The most useful thing to understand first is that this is not a noise problem. It feels like one, because the music is unwanted sound sitting behind your voice, but the tools that remove hiss and hum work in a completely different way, and they will not do this job.
Why normal noise reduction cannot do it
Noise reduction in REAPER, using ReaFIR in Subtract mode, works by learning a steady noise profile. You give it a slice of audio that is noise only, with no voice, it measures the constant hiss or hum sitting there, and it subtracts that same fingerprint across the whole recording. It works because room hiss, a fan, or mains hum stay roughly the same from moment to moment.
Music does the opposite. It changes every instant: new notes, new drum hits, new chords. There is no single steady fingerprint to capture, so there is nothing stable to subtract. Point noise reduction at music and you either barely touch it or you tear holes in the voice trying. This is why the usual hiss and hum guides do not solve it. For genuinely steady noise those methods are still the right call, and you can see them in how to reduce background noise in a podcast in REAPER and how to remove hum from a recording in REAPER. Music is simply a different problem.
What actually works: AI stem separation
Separating a voice from music that is mixed with it is a job for a trained machine-learning model. Instead of subtracting a fixed profile, these models have learned what a human voice looks like versus what drums, bass, and instruments look like, and they split a finished mix back into rough parts, usually a "vocals" stem and an "instrumental" or "other" stem. This is the same technology behind karaoke and vocal-remover tools, just used the other way around: you keep the vocal and throw away the music.
REAPER does not do this on its own. There is no stock plugin that separates a mix into stems, so the separation step happens in a dedicated tool, and then you bring the result back into REAPER to finish. Good news for the budget: the strongest option is free.
The one most people land on is Ultimate Vocal Remover (UVR), a free, open-source app for Windows, macOS, and Linux. It runs the separation on your own computer using downloadable models based on Meta's Demucs and the MDX-Net family, so there is no upload, no account, and no per-file cost. If you would rather not install anything, browser-based separators exist too, though free web tools often cap length or quality and involve uploading your audio. For a one-off clip a web tool is fine. For anything you care about, a local run through UVR usually gives you more control.
The workflow, step by step
Export the mixed file first
If your recording lives in a REAPER project, render the section you want as a single WAV. If it is already a standalone file, you are ready. Work from the highest-quality version you have. A separator can only work with what is in the file, so a compressed, low-bitrate copy gives a worse result.
Run it through a separator
In UVR, load the file, pick a vocal-separation model, and let it process. It writes out two files, roughly a vocal and an instrumental. Try more than one model if the first result disappoints, because different models handle different material better, and there is no single best one for every recording.
Bring the vocal stem back into REAPER
Import the "vocals" file onto a track. Solo it and listen closely, especially in the gaps between words where any leftover music is easiest to hear.
Clean up what is left
Separation is rarely perfect. Use a high-pass filter to clear rumble, a gentle EQ cut to reduce any obvious ghost of the music, and light noise reduction only if a steady hiss remains. Go easy. Every extra move on an already-processed file risks making the voice sound more watery.
These tools ship with several separation models and let you download more. One model might leave a faint cymbal wash, another might dull the voice. Running the same file through two or three and keeping the best output is normal, and it is the single biggest thing you can do to improve the result without spending money.
What this will not do
Separation has real limits, and knowing them saves you hours of chasing a result that is not there.
- It will not give you a clean studio vocal. Expect some artifacts: a slightly watery or smeared tone on the voice, and often a faint ghost of the music underneath, strongest where the voice was quiet.
- It struggles most when the voice and music share the same range, for example a vocal sitting right on top of a busy guitar or synth in the mids. Where they overlap, the model has to guess, and guesses leave marks.
- It cannot rebuild detail that the mix buried. If the voice was already drowned out, no tool will pull a strong, natural vocal back out of it.
- It does not clean up a phone or camera recording of music playing in a room as well as it splits a proper studio mix. Room recordings are messier to begin with, so the separation is messier too.
On a clearly produced track, a good separation can get you a usable voice for editing, transcription, a rough remix, or a quick fix. It will not sound like the voice was recorded alone in a booth. If broadcast-clean isolation is the goal and the parts overlap heavily, the honest answer is that a mixed file may not get you there.
The move that beats every plugin
If you have any access to the source, use it before you reach for separation at all. If the music and voice were ever on separate tracks, in your own REAPER project or a session someone can send you, muting or lowering the music track is a perfect result with zero artifacts. Separation only exists because that original is gone. So the order is: find the multitrack or the stems first, and only separate a finished mix when nothing else is available.
Separating a commercial song into stems does not give you any rights to use them. The tool will happily process copyrighted material, but what you are allowed to do with the result is a separate question. Keep separated stems of music you do not own for private practice, not publishing.
| Your situation | Best move | Expected result |
|---|---|---|
| Music and voice are on separate tracks | Mute or lower the music track | Perfect, no artifacts |
| One mixed file, clearly produced music | AI stem separation, then cleanup in REAPER | Usable voice, minor artifacts |
| One mixed file, voice buried or overlapping | Separation, but manage expectations | Rough, often not broadcast clean |
| Steady hiss or hum, not music | Noise reduction, not separation | Clean, this is the right tool |
If it still sounds off
- You only tried one model. Run it again with a different separation model and compare. This fixes more problems than any amount of EQ afterward.
- You over-processed the stem. Stacking heavy EQ, noise reduction, and a de-esser on an already-separated file makes it worse. Strip your effects back and add only what the voice actually needs.
- You started from a low-quality file. Separating an MP3 that was already compressed twice gives a poor result. Find the best original you have and separate that.
- It is really a one-off noise, not music. A single cough, click, or bump is a spot repair, not a separation job. See how to remove a cough or one-off noise in REAPER.
Music mixed into a voice recording cannot be removed with noise reduction, because it is not steady noise. Use free AI stem separation such as Ultimate Vocal Remover to split the file, keep the vocal stem, and finish it in REAPER, trying a couple of models and keeping the cleanest one. Expect a genuine improvement, not a perfect studio vocal, and always use the original separate tracks instead if you can find them.
Sources
- Cockos, REAPER User Guide (current version 7.79, released August 17, 2026) - ReaFIR and the stock noise-reduction workflow, which subtracts a steady sampled noise profile. reaper.fm/userguide
- Ultimate Vocal Remover (Anjok07), open-source, MIT-licensed GUI for audio source separation, cross-platform, using Demucs and MDX-Net based models run locally. github.com/Anjok07/ultimatevocalremovergui
- LANDR, overview of current AI stem-separation and vocal-removal tools and their typical strengths and limits. blog.landr.com/ai-stem-splitters