Stem splitter
How to split a song into stems: vocals, drums, bass and more
Split a finished song into vocals, drums, bass, instrumental and other stems. Learn what separation can recover, which file to upload, and how to use the results.
September 10, 20268 min read

To split a song into stems, upload the cleanest version of the finished mix to a source-separation tool, choose the parts you actually need, then download and audition each result against the original. A stem splitter can estimate vocals, drums, bass and other instruments, but it cannot reconstruct the exact multitrack session that made the record.
Know what a stem splitter gives you
A stem is a grouped audio part that begins at the same point as the full song. A vocal stem may contain lead and backing vocals; a drum stem may contain the whole kit and percussion. When all stems are aligned and played together, they should closely resemble the original mix.
Source separation works backwards from a stereo master. It estimates which sound belongs to each musical role; it does not recover the producer’s original vocal, kick, snare and synth tracks.
Choose the outputs and the best source
Start with the job, not the largest stem count. Vocals plus instrumental is enough for karaoke, practice and quick references. Separate drums and bass become useful when you need to study, edit or remix individual roles.
Use a legitimate WAV, AIFF or FLAC when possible. A good MP3 can work, but repeated lossy compression makes cymbals, consonants and reverb tails harder to assign cleanly.
Split the track and keep every file aligned
Choose Standard for free vocals and instrumental, or Studio for vocals, drums, bass, other and instrumental as 44.1 kHz WAV files. Leave the page open while the GPU processes the track, preview the result, then download individual stems or the bundle.
Every file keeps the same original start point. Import the stems at bar one rather than trimming each file at its first sound; shared timing is what makes them useful as a session.
Check the difficult edges
Solo the first vocal entrance, a quiet breakdown, a cymbal decay and a passage where bass and kick land together. These moments reveal whether reverb, distortion or overlapping frequencies crossed into the wrong stem.
Some bleed is normal. Repair only the audible problem in context: a faint hi-hat may disappear under a new beat, while an exposed vocal intro may need a short fade or carefully placed silence.
Turn the stems into a useful session
Keep an untouched copy of every stem and work on duplicates. Use gain before heavy processing, compare often with the original, and do not normalise every file independently if the original balance matters.
Use the vocal guide when the voice is the destination, or the drum guide when you need the groove. Each starts with the same split and follows one part into a different working session.
Split a song in four steps
- Choose the highest-quality source file you are allowed to use.
- Select Standard for vocals and instrumental, or Studio for five WAV stems.
- Preview exposed entrances, decays and dense passages.
- Download the stems without trimming their shared start point.
Continue with one part
- Isolate vocals for an acapellaKeep the voice aligned and clean only the edges that will be heard.
- Isolate drums without losing the groovePreserve transients, timing and the relationship with the bass.
Frequently asked questions
- How do I split a song into stems?
- Upload a song to a stem splitter, choose whether you need a simple vocal-and-instrumental split or separate vocals, drums, bass and other parts, then download the results. CPL OF MNGS Standard makes vocals and instrumental for free; Studio returns five WAV stems.
- Can a stem splitter recover the original multitrack recording?
- No. A stem splitter estimates the sources inside a finished mix; it does not recover the exact tracks used in the original session. Reverb, distortion, layered parts and shared frequencies can leave some bleed or artifacts.
- What audio file should I use for stem separation?
- Use the highest-quality legitimate source you have, ideally WAV, AIFF or FLAC. A good MP3 can still work, but repeated compression gives the model less clean information and can make cymbals, vocal edges and reverb tails harder to separate.





