Stem splitter guide: when to choose two, four or six tracks
A stem splitter estimates several musical groups from a finished song. Choose the output groups you actually need: two tracks for voice and backing; four when drums and bass need their own controls; six when guitar or piano is also your target. More outputs give you different editing options, not a guarantee of better separation. Both stem categories and quality depend on the model.
1. Choose two, four or six stems for your goal
In a recording project, stems can be rendered groups such as all backing vocals or all drums. Individual microphone or instrument tracks may still exist inside the project. An AI stem separator starts with a mixed recording and estimates groups; it does not reopen the original session. If original session exports are available, ask for those before trying to reconstruct them.
- Two stems: Vocals and Instrumental. Useful for karaoke, acapella extraction and comparing the vocal/music boundary.
- Four stems: Vocals, Drums, Bass and Other. Useful for rhythm-section study or adjusting the relationship between those groups.
- Six stems: Vocals, Drums, Bass, Guitar, Piano and Other. Useful when guitar or piano needs a separate estimate.
- Target instrument: a named instrument and a remaining-song output, where that processing option is available. Useful when you want an isolated part or a ready-to-audition backing with that part reduced.
These are the intended layouts in the configured uvr online modes, not a claim that every sound will land perfectly in its named file. Other products also define output categories and collect unselected material into a residual group; check the actual list, not just the advertised track count.
2. “Other” is a real output, not an error bin
In a four-stem result, guitar, piano, strings and synths can all share Other. You cannot get a clean piano file by renaming Other to Piano. A six-stem model has more categories, but still does not promise one file per violin, guitar take, keyboard or backing singer. Keep Other when downloading a complete set, because discarding it can remove much of the arrangement.
3. Check the result before rebuilding your mix
After the full result is ready, compare the difficult passage with the original. Check both the wanted detail and unwanted leakage before choosing a track.
For example, a piano that seems absent from its own track might still be audible in Other. Listen across the whole output set before deciding that a note was not present in the original. For practice, a small amount of spill may be acceptable if it preserves the rhythm you are learning. This is an editorial listening test, not evidence that any particular model wins on your recording. Isolated listening can help reveal musical relationships that are hard to follow in a full mix.
To remove drums, bass, guitar or piano, use the corresponding target mode when available and inspect its remaining-song file. Alternatively, download a multitrack set and combine the wanted stems in an editor. Player mute and volume controls help audition a balance; they do not by themselves establish that a custom rendered mix has been saved. Keep all files at the same start and check the final combined result.
Save the model, mode and source filename with your downloads. If you later compare another splitter or model, use the same source passage and similar playback level. A track count is an output description, not a quality score.
4. Frequently asked questions
Does six-stem separation extract every instrument?
No. It estimates named groups. Multiple guitars, several pianos or strings may still be grouped or leak between outputs.
Which mode should I use to remove one instrument?
Try its target-instrument mode if available. For more control, use a multitrack result and combine the remaining files in an audio editor.
Why is there music in Other?
Other holds material outside the explicit categories and can include imperfectly assigned sounds. It is part of the result, not automatically disposable noise.
What are vocal stems and music stems?
They are audio parts grouped by content. The vocal stem holds an estimate of singing; music stems may divide drums, bass and other groups. In an AI stem splitter these are reconstructed estimates, not a recovery of each studio recording.
Is a free stem splitter enough to separate instruments from a song?
It depends on the available target categories and your use. Check current access and processing limits, then evaluate the wanted instrument. A free or paid label alone does not tell you whether a specific recording separates well.