Online stem splitter

Split any song into stems online

Get separate vocal and instrumental tracks from any song in the browser. For deeper 5-stem splits (vocals, drums, bass, piano, other), continue on the mobile app.

  • stem splitter
  • remix and DJ prep
  • practice stems
  • free browser test
Unmix splitting a song into vocals, bass, drums, chords, and other stems

Drop a song here — or tap to try it on your track

Free, in your browser. No signup. MP3, WAV, FLAC, M4A, OGG, or video.

Choose a file
Want 5-stem (drums, bass, piano)? iOS App Android App

Who uses a stem splitter

Different people need different things from the same tool.

DJs

Vocal and instrumental layers for mashups, transitions, and live sets. Quick 2-stem split in the browser is usually enough for testing.

Producers

Reference stems for sampling, arrangement study, and remix prep. Most chain separation with DAW cleanup for production use.

Teachers and students

Practice-ready stems for focused listening, drills, and play-along. Remove the instrument you want to practice.

2-stem vs 5-stem separation

The browser and mobile apps offer different levels of separation. More stems means more control but also more chances for bleed between instruments.

Browser (free)Unmix App
StemsVocals + AccompanimentVocals, Drums, Bass, Piano, Other
PriceFree, no accountOne-time purchase
ProcessingOn your device in browserOn your device, fully offline
Artifact levelLower — only one split decisionHigher — 5-way split means more bleed
Best forQuick test, karaoke, quality checkProduction, practice, individual instrument control

What to expect from each stem

Not all stems separate with equal quality. Here is how they typically rank.

Vocals: best results

Vocals are the easiest to separate. Centered, dry vocals in clean mixes come out the cleanest.

Drums: usually good

Kick and snare reduce well, but cymbal and hi-hat energy sometimes leaks into other stems.

Bass and piano: hardest

Bass shares low-end with kick drums. Piano shares harmonic space with guitars and synths. Expect partial isolation on these.

Remove a specific instrument

Stem workflows for DJs and producers: live decks vs pre-prep

Serato Stems, Virtual DJ Stems, and Traktor Pro Plus do separation in real-time on the deck. When they fail — or when you need cleaner output — these are the scenarios where pre-extracting with a dedicated tool wins.

Live stem DJing: when the deck isn't enough

Serato Stems and Virtual DJ Stems separate in real-time but use smaller, faster models that struggle on 808-heavy hip-hop and dense rock. Pre-extract with Unmix or the Unmix app, drag the separated stems into your deck, and trigger them individually. Especially useful for mashups where you need a vocal from one track and a clean drumline from another without real-time artifacts ruining the transition.

Sample-pack building: BPM_key naming saves hours later

When building a sample library from separated stems, use a folder structure like /drums/120bpm/ and name files 120_Am_KickLoop.wav. Tools like Rekordbox and Ableton Live can auto-warp and key-match if the metadata is right. Run BPM and key detection (Mixed In Key, Rekordbox's analyzer) after extraction, not before — separated drums occasionally confuse BPM detectors.

Model reality: what each separator can actually output

Spleeter (Unmix's browser model) outputs 2 or 5 stems: vocals, drums, bass, piano, other. Demucs v4 hybrid-transformer goes to 6 by separating guitar from 'other'. MDX-Net models top 4 stems but with fewer artifacts. If you specifically need a guitar stem, browser Spleeter won't get you there — use Demucs via MVSep or the Unmix app's higher-tier model.

Test it on your own track

Upload any song and hear the separated stems in seconds. Free, no account needed.

Tips for better results

Use WAV for production stems

If the stems are going into a DAW, start with lossless input. The quality difference is noticeable on bass and piano stems.

Test with 2-stem first

The browser's free 2-stem split tells you quickly whether the track is a good candidate for deeper separation on mobile.

Layer stems from different tools

Some users get the best drum stem from one tool and the best vocal from another, then combine in their DAW.

FAQ

Why do results vary between songs?

Separation quality depends on mix density, effects, stereo placement, and source file quality. Sparse mixes with well-separated instruments give the best results.

What is the difference between 2-stem and 5-stem?

The browser splits into vocals and accompaniment. The mobile apps go further with drums, bass, piano, and other as separate stems.

Which stem is the hardest to isolate?

Bass and piano are typically the hardest because they share frequency ranges with kick drums and guitars respectively.

Do I need multiple tools?

For most tracks, one tool is enough. On difficult material, advanced users sometimes compare outputs from 2-3 tools.

What's the difference between Unmix and Serato Stems or Virtual DJ Stems?

Serato, Virtual DJ, and Traktor Pro Plus do separation in real-time on the deck. Speed is their advantage; quality is the tradeoff — they use smaller models that struggle with dense mixes. Unmix and similar pre-processing tools let you run heavier models (Spleeter/Demucs/MDX) offline, export cleaner stems, then drop them into your deck of choice. Use deck-stems for most tracks; pre-extract the difficult ones.

Can I get a guitar-only or synth-only stem?

Not from the browser version. Spleeter's 5-stem model separates vocals, drums, bass, piano, and 'other' — guitars and synths both land in the 'other' bucket. For a dedicated guitar stem, use Demucs v4 (via MVSep.com or the Unmix mobile app), which separates guitar as its own stem. Expect bleed between guitar and piano when both are harmonically layered.

How do I batch-process stems for a full DJ set?

The browser tool runs one file at a time to respect memory limits. For batch work across a 20-40 song setlist, use the Unmix mobile app (iOS and Android) which queues extractions and can run overnight on charging. Alternative: a local install of Demucs via Python handles batches of hundreds but needs a capable GPU (Nvidia RTX 3060 or better).

Explore more tools