Audacity 4 has no vocal remover effect. The built-in route is the classic center-cancel trick, done with three commands it still ships. Split the stereo track into two centered mono tracks, apply Effect ▸ Special ▸ Invert to one of them, and export the pair as one file. Anything panned dead center, which is where most lead vocals sit, cancels out. For a cleaner result with fewer side effects, the free OpenVINO Music Separation effect does the job with an AI model, and it needs Audacity 3. A stem-separation plugin from MuseHub is the third route.
This guide walks through all three routes with the exact Audacity menus, adds the no-install option in the browser, compares them in one table, and says plainly what each one cannot do. The same steps isolate the vocal instead of removing it.
Tested on Audacity 4.0.1 on macOS Sequoia; the OpenVINO route was tested on Audacity 3.7.9 on Windows 11, where its two screenshots were taken, and on macOS.
Audacity and the vocal-separation tools
What you need
- A true stereo file. Every route except the AI ones depends on the vocal being in the center of a real stereo mix. A mono file, or a stereo file with identical left and right channels, gives the center-cancel trick nothing to work with.
- Audacity 4 for the built-in route, Audacity 3 for the AI route. Both are free and both install from the Apps section of MuseHub; they can live side by side on one computer. How to use effects and plugins in Audacity covers installing and validating plugins in Audacity 4.
- A few minutes for any route. Music Separation runs on your own processor. Four stems from a 30-second excerpt took about 10 seconds on a CPU, and the first run is slower while the model loads.
Route A: cancel the center with Invert (built in, Audacity 4)
This is the trick every “vocal remover” was built on before AI arrived. A mixed song puts the lead vocal at equal level in the left and right channels. If you flip the polarity of one channel and add the two together, everything identical on both sides sums to zero, and the vocal disappears. Everything panned left or right stays.
- Import the song with File ▸ Import… (Ctrl+Shift+I / ⌘⇧I). It lands as one stereo track.
- Click the ⋯ button on the track’s control panel, at the left of the track, and choose Split stereo to center mono. You now have two mono tracks, both centered and both named after the original.
- Select the lower track by click-dragging across its whole length. Ctrl+A / ⌘A selects every track, which would flip both.
- Choose Effect ▸ Special ▸ Invert. There is no dialog; the waveform flips upside down immediately. Played alone it sounds identical, which is expected.
- Press play. Audacity sums the two tracks on playback, and the vocal falls far back. If it did not drop at all, the file is mono or the vocal was not centered; see “What to expect” below.
- To get one file, export with File ▸ Export audio…, which mixes the two tracks on the way out and bakes in the polarity flip. Audacity 4.0.1 has no mix-down command, so the export is the mix-down step.
The result is mono, because both channels are now the same signal. Whatever else sat in the center goes with the vocal, usually the kick drum, the bass and the snare. That is the hollow, thin sound people describe after a phase cancel, and no setting fixes it, because the trick cannot tell a voice from a bass guitar.
On Audacity 3. Older guides send you to Effect ▸ Special ▸ Vocal Reduction and Isolation. Audacity stopped shipping that effect in version 3.5.0, but it is still a free Nyquist plugin you can download and enable; its page in the Audacity manual has the details. It runs the same center cancel with a Strength slider and a frequency range, so it also needs a true stereo file.
- Run Analyze first. It reports how alike the two channels are. A correlation near 0 means the channels share almost nothing, and the effect will not work on that file; go to Route B.
- Set Low Cut by voice. The manual’s default of 120 Hz suits most lead vocals; it suggests about 170 Hz for a female voice.
- On Audacity 4. Audacity 4 loads Nyquist plugins, so you can try the same download there. If it does not show up under Effect, Route A does the same job with no download.
- “Effect not found” on a Mac. If the plugin, or any built-in effect, shows up but will not run, the fix is in the FAQ below.
Route B: OpenVINO Music Separation (AI, Audacity 3)
Music Separation uses a machine-learning model that has learned what a voice, a drum kit and a bass sound like. It pulls the vocal out by recognizing it, wherever it sits in the stereo field. It is part of OpenVINO AI Tools, free from Muse Group and Intel, and it processes on your own computer with nothing uploaded.
The catch is the version. The plugin supports Audacity 3.7.4 through 3.7.9 and does not run in Audacity 4 yet, so install Audacity 3 from MuseHub alongside Audacity 4. It runs on Windows and macOS, including Apple Silicon Macs, but not on Windows on Arm. On a Mac it uses the CPU only and is much slower: our first four-stem run of a 30-second excerpt took about three minutes there, against about ten seconds on Windows. How to edit audio with AI in Audacity covers the install.
- In MuseHub, install OpenVINO AI Tools. On the model page of the installer it opens, include Music Separation; Install recommended models covers it. The models are large, so leave out the ones you will not use.
- Open Audacity 3 and import the song.
- Select the track and choose Effect ▸ OpenVINO AI Effects ▸ OpenVINO Music Separation.
- Pick the Separation Mode. (2 Stem) Instrumental, Vocals gives you a vocal track and an instrumental track. (4 Stem) Drums, Bass, Vocals, Others gives four, which is the choice when you want to keep the drums and drop the vocal and the bass.
- Choose the OpenVINO Inference Device, CPU or GPU (or NPU on recent Intel chips), and click Apply. The first run is the slow one, because the model loads; later runs are faster.
- The stems arrive as new tracks under the original, named after it: song-Drums, song-Bass, song-Other Instruments and song-Vocals for a file called song. Mute the vocal track and the original, then export for a backing track, or solo the vocal for an a cappella.
If it fails. “Music Separation failed” with an empty device list can follow a reinstall of OpenVINO AI Tools while Audacity 3 was open. Restart Audacity and the CPU shows up again. For the rest of the plugin, including transcription and noise suppression, see how to edit audio with AI in Audacity.
Route C: a stem-separation plugin
MuseHub has no vocal-remover plugin; its vocal remover runs in the browser, below. The one catalog product that separates stems is Neutone FX, a free VST3 and AU plugin from Neutone that loads AI models and runs them on the audio passing through it. Its model list includes Demucs.StemSeparation by Bogdan Teleaga, which splits the input into bass, drums, vocals and other, with a knob per stem.
It runs as a realtime effect in Audacity 4; these steps were tested on macOS.
- Install Neutone FX from MuseHub and start Audacity 4, which validates the plugin.
- Select the song’s track, press E to open Realtime effects, click Add effect and choose VST3 ▸ Neutone ▸ Neutone FX.
- In the plugin window, pick Demucs.StemSeparation from the model list and click Use. It downloads inside the plugin, with no account or sign-in.
- The stems are the knobs A to D under custom settings: A is bass, B drums, C vocals and D other, and each name shows on hover. The model loads with only C up, which isolates the vocal. To remove it, turn C down and A, B and D up. If a knob does not respond at first, click it once.
- Play to check the result, then export; the realtime effect is written into the file on export.
On a pop song the vocal was mostly gone, with some audible bleed and the music intact. It is the route to try when you already work in Audacity 4 and do not want a second Audacity for one job.
No install: the MuseHub Vocal Remover in the browser
If you only need the instrumental and do not want to set up Audacity 3 or a plugin, the MuseHub Vocal Remover does the separation online and for free. Upload an MP3, WAV, FLAC, M4A or OGG file of up to 1 GB or 20 minutes, preview the result and download the instrumental. Its sister tools take the other jobs: the Voice Extractor keeps the vocal and drops the rest, for an a cappella, and the Stem Splitter gives you vocals, drums, bass, guitar, piano and the rest as separate files. The trade against the local routes is the upload: the song leaves your computer, while Routes A to C run on your machine.
Outside MuseHub, Moises and LALAL.AI are the online services most people land on; both meter minutes or charge a subscription.
The routes compared
| Route | Audacity version | Quality | When to use it |
|---|---|---|---|
| A: center cancel with Invert | Audacity 4 (built in); Audacity 3 via the downloadable Vocal Reduction and Isolation plugin | Fair: vocal reduced, not gone; mono result; kick, bass and snare go with it; reverb tails and doubles stay | A quick practice or karaoke track from a clean, centered stereo mix; no install |
| B: OpenVINO Music Separation | Audacity 3.7.4–3.7.9 only | Good: real stems, stereo kept, works on doubles, reverb and mono files; some bleed on dense, loud masters | Backing tracks, a cappellas, sampling, when you want the drums and bass intact |
| C: Neutone FX with Demucs | Audacity 4 (realtime VST3 / AU; tested on macOS) | Good, a step below B: realtime model, more bleed on busy mixes | Staying in Audacity 4; quick stem previews; remixing without leaving the session |
| MuseHub Vocal Remover (online) | Any; the download is imported like any file | Good: AI separation; files up to 1 GB or 20 minutes | No install, a one-off backing track; skip it when the song must not leave your computer |
What to expect
Every route reduces the vocal; none erases it. Separation is reconstruction, and what survives depends on the source.
- Center-panned lead vocal, dry mix: Route A gets close, Routes B and C get very close.
- Reverb and delay on the vocal: the tail is spread across the stereo field, so Route A leaves a ghost singing along. B and C handle most of it.
- Doubled or harmony vocals panned wide: invisible to Route A. B and C treat them as vocals.
- Live recordings and mono files: Route A does nothing useful. B and C still work, with more bleed.
- Loud, heavily compressed masters: all three leave a watery shimmer around where the vocal was. B is the cleanest.
Common mistakes
- The file is mono. Two identical channels cancel to silence, or a stereo file made from a mono source cancels nothing. Check the track: a stereo track shows two waveforms that differ.
- Inverting both tracks. Ctrl+A / ⌘A selects every track, and both flipped is the same as neither flipped. Invert exactly one.
- Splitting to L/R mono, then wondering why nothing cancels. The two tracks must both play from the center. Use Split stereo to center mono, or drag both pan knobs to the center.
- Starting from an MP3. Encoding artifacts differ between channels, so the cancel is incomplete and the AI model hears noise that was never in the recording. Use WAV or FLAC when you can.
- Expecting a released-quality instrumental. For practice, karaoke and sampling, yes. For a release, listen critically on headphones for bleed.
Tips that improve the result
- Normalize afterwards. A canceled mix is quieter. Effect ▸ Volume and compression ▸ Normalize to about −1 dB brings it back up without clipping.
- Gate the tails. After Route A, Effect ▸ Noise removal and repair ▸ Noise gate can silence the faint reverb ghosts between phrases. Set the threshold just above the ghost and no higher, or the quiet parts of the music start stuttering.
- Keep the original track. Mute it rather than deleting it, so you can A/B the result and go back.
Vocals on the left, music on the right
For singing practice, some people want the voice in one ear and the backing in the other. Separate the song first with Route B or C, so you have a vocal track and an instrumental track. Turn the vocal track’s pan control fully left and the instrumental’s fully right, then export with Channels set to Stereo. The usual reason this fails is a mono export, which folds both sides back into the middle.
Audio in the screenshots: “Pop Vocal” by The_Mountain on Pixabay, used under the Pixabay Content License.
FAQ
Next steps
With the backing track ready, record your own vocal over it in Audacity, then tune it with the free Graillon 3. For the plugins that shape a vocal once it is recorded, the best vocal plugins covers the chain from EQ to reverb, and the best AI plugins covers stem separation inside a DAW. Get Audacity, OpenVINO AI Tools and Neutone FX in one place: download MuseHub free.
