Free · No account · Unlimited splits

Split any song into isolated stems

Upload a track and get separate files for vocals, drums, bass, piano and everything else — ready to preview in the browser and download.

Useful for karaoke backing tracks, remixes, transcription, practice loops and rebalancing a mix you can no longer re-open.

Your file is uploaded to our processing server, separated, and then deleted — nothing is stored permanently and no account is needed. See the privacy policy for details.

What VocaSplitter does

VocaSplitter takes a finished stereo recording and reconstructs the individual parts inside it as separate audio files. Upload one song and you get back a vocal track with no band behind it, a drum track with nothing else on it, a bass line on its own, and so on — each as a normal audio file you can play, edit or import into any DAW.

It runs a source separation model of the Spleeter family on our server. That matters for one practical reason: separation is a genuinely heavy computation, and doing it server-side means the quality does not depend on how fast your laptop or phone is. A three-minute song typically takes one to three minutes depending on how busy the queue is.

The tool is free and has no upload limit, no sign-up and no watermark on the output. It is built and maintained independently, and running costs are covered by advertising on the written guides rather than by charging for separations.

How it works, in three steps

Upload your track

Drop in an MP3, WAV, M4A, OGG or FLAC file. Nothing is required beyond the file itself — no account, no email, no watermark on the result.

Pick a separation mode

Two stems for a clean vocal-and-instrumental split, three to also pull the drums out, or five for vocals, drums, bass, piano and everything else.

Preview, then download

Play each stem in the browser before committing to anything. Download individual files, or take the whole set as a single ZIP.

For the detail behind each step — what the model is actually doing, and why the source file you choose matters more than any setting — read how VocaSplitter works.

Choosing a separation mode

More stems is not the same as better quality. Every additional stem is another decision the model has to get right, so the mode with the fewest classes that answers your question will usually sound best.

Mode	Files you get	Best for
2 stems	Vocals, Instrumental	Karaoke tracks, acapellas, removing a voiceover
3 stems	Vocals, Drums, Instrumental	Drum practice, beat transcription, DJ edits
5 stems	Vocals, Drums, Bass, Piano, Other	Remixing, rebalancing a mix, learning one instrument

Read the full decision guide for how source material changes the answer.

What people use it for

Karaoke and backing tracks

A two-stem split gives you an instrumental you can sing over. Our karaoke guide covers the level and ghosting work that makes a track usable in a real room.

Remixing and DJ edits

Loop an isolated drum stem to extend an intro, drop the beat for a breakdown, or build a mashup from an acapella and a separate instrumental.

Learning parts by ear

Hearing a bass line without the kick drum on top removes the hardest step in transcription. Mute your own instrument and play against the original rhythm section.

Teaching and rehearsal

Hand each section of an ensemble a reference of their own part, or show a student the difference between what is on the record and what they are playing.

Video and podcast editing

Pull a music bed out from under a voiceover, or recover a clean narration track from a mix where the session files are long gone.

Rebalancing a finished mix

When the multitrack is lost, separated stems let you bring a buried vocal forward or pull down a piano that sits too loud, then bounce a new version.

What to realistically expect

Separation is an estimate, not a recovery of the original session. When a song was mixed down, dozens of sources were summed into two channels and the information about what came from where was genuinely lost. The model reconstructs a very well-informed guess. Understanding that explains almost everything you will hear.

Sparse recordings separate beautifully — a voice with a guitar, a soul record, most pre-2000 rock. Dense, heavily-limited modern productions with layered synths are much harder, and you will hear a faint shadow of the vocal in the instrumental. Songs with heavy autotune or a hard-doubled lead are harder still, because the voice no longer resembles the natural singing the model learned from.

Source quality matters more than any setting. A 128 kbps MP3 has already had most content above roughly 16 kHz discarded by the encoder, and the model will faithfully separate a song with no top end. Feed it a WAV, a FLAC, or at minimum a 256 kbps MP3.

For practice, transcription, karaoke, DJ edits and sample sourcing, the quality is well past the point of being useful. For commercial release of an isolated vocal it usually is not — you can hear the process on headphones if you listen for it. Our guide to fixing common artifacts covers what can and cannot be repaired afterwards.

Supported files

MP3, WAV, M4A, AAC, OGG, FLAC and AIFF.
Stereo or mono. Stereo separates better, since the model can use position as a cue.
Any sample rate. Output comes back at the sample rate you uploaded.
If a file is rejected, re-export it as a 44.1 kHz WAV — that repairs most container-level problems.

Your files and your privacy

Uploads are sent to our processing server over HTTPS, separated, and then deleted. We do not keep a library of what you upload, we do not train models on it, and we do not pass it to anyone else.

Because files are removed after processing, download your stems during the session rather than bookmarking the links. Full detail is in the privacy policy.

Guides and tutorials

Longer write-ups on how separation works and how to get real work out of it.

Technology · 9 min read

How AI Stem Separation Actually Works

A plain-English explanation of how a neural network pulls vocals, drums and bass out of a finished stereo mix — spectrograms, masking, and why artifacts happen.

Tutorials · 8 min read

How to Make a Karaoke Track From Any Song

A complete walkthrough for producing a clean, singable backing track — choosing the right split mode, dealing with leftover vocal ghosting, and getting the level right for a live room.

Guides · 6 min read

2, 3 or 5 Stems: Choosing the Right Separation Mode

More stems is not better quality. A decision guide covering what each separation mode actually gives you, when the extra classes help, and when they quietly destroy your audio.

Troubleshooting · 9 min read

Fixing the Most Common Stem Separation Artifacts

Watery vocals, hi-hat bleed, missing bass punch and vocal ghosting — what causes each artifact and the specific repair that actually works.

Tutorials · 7 min read

Getting Separated Stems Into Your DAW Without Losing Sync

Sample rates, alignment, tempo detection and gain staging — the practical steps for turning downloaded stems into a session you can actually work in.

Guides · 8 min read

Stem Separation and Copyright: What You Can and Cannot Do

Separating a song creates a derivative work. A practical, non-legalistic look at private use, covers, remixes, sampling, karaoke and monetised video — and where the lines actually fall.

Guides · 7 min read

How Musicians Use Isolated Stems to Learn Songs Faster

Practical practice methods built on separated audio — transcribing a buried bass line, drilling a groove against the original drummer, and building ear training that actually transfers.

Please only process audio you have the right to use

Splitting a commercial recording creates a derivative work. Doing that for private study or practice is uncontroversial in most places; publishing, performing publicly, or monetising the result generally needs permission from whoever controls the recording and the composition.

We explain where the lines fall in stem separation and copyright, and the terms of use set out what we ask of you.