Make an instrumental practice track with Audacity’s AI separation

A rehearsal recording can be more useful when you can turn down the lead vocal and sing over the accompaniment. Audacity’s OpenVINO Music Separation effect offers a route: it estimates separate vocal and instrumental tracks from an existing mix. The processing runs locally on the computer.[4][5]

Start with a recording you made or have permission to use. For a first attempt, choose a clearly sung section with accompaniment and a few pauses between phrases. Those pauses will help you hear what the separation has left in each track.

Match the editor and plugin first

This is a documentation-based workflow for 64-bit Windows, checked on 18 October 2026; we have not run this plugin ourselves. Use the explicitly documented pair of Audacity 3.7.1 and Intel’s OpenVINO AI Plugins v3.7.1-R4.2. The release notes also confirm compatibility with 3.7.3. Audacity’s current download page warns that the plugins are incompatible with the latest editor, so “install the latest of both” is not a reliable instruction.[2][1]

1. Open the linked v3.7.1-R4.2 release page. It links to the matching Audacity release and the Windows plugin installer.
2. Install the matching editor, then run the plugin installer against that Audacity installation. Include the Music Separation models and check that the installer targets the same Audacity folder.
3. Open Audacity. In Edit → Preferences → Modules, set mod-openvino to Enabled, confirm, then close and reopen Audacity.

Try one passage before processing the song

1. Import the recording into a new project, then select a short passage on its audio track.
2. Open the Effect menu and choose OpenVINO Music Separation. Select 2-Stem for Instrumental and Vocals. Choose CPU if it is offered and keep the advanced settings at their defaults for this first pass.
3. Click Apply. The first use on a device includes loading and compiling the model. When processing finishes, two additional tracks should appear with names identifying their parts.

Solo each output and listen to the phrases and the gaps. Is the accompaniment clear enough to follow? Is the vocal still distracting in the instrumental track? Listen for softened attacks, missing instrumental detail or wavering textures. A practice track needs to support the exercise; it need not sound like a studio multitrack master.

If you want separate drums or bass instead, 4-Stem produces Drums, Bass, Other Instruments and Vocals. It does not create a separate track for every instrument.[4]

Export the accompaniment you actually heard

When the short passage is useful, repeat the process in a fresh project with the whole original recording selected. This keeps the final job separate from the trial clips.

1. Mute the original mixed recording and the new Vocals track. Leave Instrumental audible and play the project to confirm the accompaniment.
2. Choose File → Export Audio. If asked, choose export to your computer. Give the file a distinct name, such as rehearsal-instrumental.wav.
3. Select WAV and set Export Range to Entire Project. This range excludes muted tracks. Current Selection has different behaviour and can export selected tracks even when they are muted, so do not carry that setting over accidentally.
4. Export, then open the saved file in a separate player. Check that the beginning, ending and vocal balance match the intended practice track.

Keep the Audacity project if you expect to make another practice version. When choosing a finished export, listen for the beat, the harmonic changes and the moment of your next entry. An accompaniment that preserves those cues can be useful in rehearsal even when close listening reveals traces of the original vocal.

Report an error or suggest a correction