About VocaSplitter

Why this exists

Source separation research became genuinely usable for ordinary music around the time Spleeter was published, and almost immediately the useful implementations moved behind paywalls, credit systems and mandatory sign-ups. That is a reasonable business decision — the computation costs real money — but it put a tool that is most valuable to people with no budget out of reach of exactly those people.

The people who benefit most from stem separation are students learning parts by ear, teachers preparing rehearsal material, singers who need a backing track for one evening, and producers trying to rescue a mix whose session files are lost. None of those are situations where paying per song makes sense.

So VocaSplitter does the separation, gives you the files, and asks for nothing. No account, no email address, no credit counter, no watermark, no upload cap.

How it is paid for

Running the separation server is the main cost. That is covered by display advertising, which appears on the written guides and explanatory pages rather than in the middle of the tool itself. There are no ads on the processing screen or between you and your download.

We would rather write something worth reading and put an ad alongside it than interrupt the thing you came here to do. If the guides are useful to you, that arrangement works.

How the separation works

Your file is uploaded to a processing server that runs a source separation model from the Spleeter family. The model converts the audio into a spectrogram, estimates how much of the energy at each point in time and frequency belongs to each instrument, and reconstructs a new audio file per stem from those estimates.

Processing happens server-side rather than in your browser. That is a deliberate choice: separation is heavy enough that an in-browser implementation would either be very slow or would have to use a much smaller, worse model, and the result would depend on whatever device you happen to be using.

There is a longer, non-technical explanation of the whole process in how AI stem separation actually works.

What we do with your audio

Uploaded files are used for one thing: producing your stems. They are deleted after processing. We do not keep a library of submissions, we do not train models on what you upload, and we do not share files with third parties.

Because of that, download links are short-lived. Save your stems during the session rather than bookmarking a URL to come back to. The privacy policy sets out what is and is not collected in full.

What we are honest about

Stem separation is an estimate, not a recovery of the original multitrack. Sparse recordings come out cleanly; dense, loud modern productions leave audible traces of the vocal in the instrumental. No tool available today avoids this, and we would rather tell you that up front than have you spend an hour wondering what you did wrong.

Where results are imperfect, the artifact troubleshooting guide explains what each problem is and which repairs actually help — including several popular ones that make things worse.

Who it is for

• Musicians working out parts by ear, or practising against an original rhythm section.
• Singers who need a backing track and do not want to buy a karaoke version.
• Producers and DJs building edits, extending intros, or rebalancing a mix they can no longer re-open.
• Teachers preparing rehearsal and demonstration material for students.
• Video and podcast editors separating narration from a music bed.

Getting in touch

Bug reports, feature requests, corrections to the guides, rights enquiries and general feedback all go to vocasplitter@gmail.com. There is more detail on the contact page, including what to include when reporting a file that will not separate.