AI Stem Splitter Online

Upload a song and the AI stem splitter returns vocals, accompaniment, bass and drums as separate cloud tracks, or run Demucs in your browser to preview, mute and download.

Hear a demo

Rights and licensing

Credits are charged only when a task succeeds. If it fails, they are returned automatically.

Loading your tasks…

Hear it before you upload

A 30-second song made for this site, already separated into tracks. No sign-in needed. Solo or mute any track and play them in sync.

  • Vocals
  • Drums
  • Bass
  • Other

Demo song and tracks: generated by SampLab Pro with AI (Mureka through WaveSpeed) and separated with the Local 6 tracks engine, for demonstration only. No licence is granted to reuse them. Rights and licensing

Where to go next

Stem separation for specific jobs

AI Stem Splitter features

  • A maple snare drum with brushed nickel lugs and a pair of wire brushes

    Four separate stem tracks

    The AI stem splitter returns vocals, accompaniment, bass and drums, not just filtered frequency ranges.

  • A gold alto saxophone with its mouthpiece cap on, resting on cream linen

    Preview and mute

    Listen to each stem from the AI stem splitter and mute what you do not need.

  • A compact analog synthesizer with wooden end cheeks and cream keys

    Download one or all

    Save a single stem, or take all four together as a ZIP.

  • A honey-colored violin and a horsehair bow beside a small cake of resin

    Straight to MIDI

    Send one stem to WAV to MIDI for cleaner notes than a full mix gives.

Who uses AI Stem Splitter

  • A kick-drum beater, a compact guitar amp, a bass-amp cabinet and a silver vocal microphone

    Practice along

    Mute the vocals or drums with the AI stem splitter and play or sing over the rest.

  • An empty studio corner with a gobo panel, a music stand and a muted floor tom

    Remixes and mashups

    The AI stem splitter pulls out the vocals or the accompaniment so you can use them in your own arrangement.

  • A mixing console close-up with four group faders raised equally

    Study a mix

    Listen to the bass and drums on their own, split by the AI stem splitter, to hear how a track is built.

  • Four unlabeled tape reels in cream, sage, rust and charcoal in a row

    Prepare for transcription

    Send a single stem to WAV to MIDI for cleaner results than a full mix.

Why choose AI Stem Splitter

  • Credits only for finished results

    A task is charged only when it succeeds. If it fails, your credits go back to your balance.

  • Real stems, not a filter

    The AI stem splitter separates vocals and instruments. For plain low, mid and high ranges, use Frequency Band Splitter.

  • Built for what comes next

    Mute the vocals to practice, or send a track from the AI stem splitter on to MIDI conversion.

  • Your results stay private

    What AI Stem Splitter creates stays in your account until you download or save it.

Explore more tools

Every SampLab Pro tool is one click away.

How to use AI Stem Splitter

  1. Coral headphones and a silver microphone resting on a plain kraft paper sleeve
    1

    Add your audio

    Upload a file or pick a recording from your library.

  2. A microphone, coral headphones, two drumsticks and a ukulele set far apart on a wooden desk
    2

    Separate

    Press the button and the AI stem splitter returns four tracks: vocals, accompaniment, bass and drums.

  3. Coral headphones on a glowing speaker, beside a kraft folder and a drumstick
    3

    Play, download or continue

    Preview and mute stems, download one or all four as a ZIP, or convert a stem to MIDI.

How the AI Stem Splitter works

This AI stem splitter isolates individual tracks from a finished audio mix. For fast four-stem results, our cloud engine separates songs into vocals, accompaniment, bass, and drums on remote servers. When you want complete privacy, the tool runs the open-source HT-Demucs model (Demucs v4, htdemucs_6s) directly in your browser to extract six stems—drums, bass, vocals, guitar, piano, and other—without your audio ever leaving your device.

Separation is not perfect because instruments share overlapping frequencies in a mix. On cloud separations, the accompaniment track contains everything that is not the vocal, meaning it already includes the bass and drums. For local Demucs processing, piano is the weakest of the six separated stems. Both options deliver separate tracks you can preview, mute, download, or send to WAV to MIDI for note transcription.

A microphone, coral headphones, a small ukulele and two drumsticks spaced on a wooden desk

What the AI Stem Splitter gives you

The AI stem splitter takes one song and returns four audio tracks: vocals, accompaniment, bass and drums, or six tracks when using local Demucs separation. The accompaniment holds everything that is not the vocal, so it already contains the bass and drums and overlaps with those tracks. Separation is not perfect and can leave faint traces of other sounds.

A wooden audio meter box with abstract LED level bars in cream and sage

AI Stem Splitter or Frequency Band Splitter

Frequency Band Splitter applies filters to give low, mid and high ranges in your browser. A guitar and a snare that share a range stay together, so it does not isolate instruments. Use the AI stem splitter when you need the voice or the drums alone; it needs an account, and a separation costs 9 credits/started min (paid min. 36).

Four small speaker cones in cream, sage, rust and charcoal

Getting better results from the AI Stem Splitter

A clean, full-quality recording with a clear vocal gives the model more to work with than a crowded mix. Free accounts can use files up to 50 MB and 1 minute long; paid plans up to 200 MB and 15 minutes. If you only need one part of a song, trim it first with Audio Cutter, and send a single stem to WAV to MIDI for notes; that uses no extra credits.

Ready to try AI Stem Splitter?

Add a song at the top of the page and this AI stem splitter returns four separate tracks. A separation costs 9 credits/started min (paid min. 36), charged only when it succeeds.

Open AI Stem Splitter

AI Stem Splitter FAQ

How many credits does the AI stem splitter use?

Each task costs 9 credits/started min (paid min. 36). Credits are charged only when a task succeeds; if it fails, they go back to your balance.

What does an AI stem splitter do?

An AI stem splitter takes a finished song and separates it into vocals, accompaniment, bass and drums. You can preview, mute and download each stem.

Which files can I use?

Common audio formats. Free accounts can use files up to 50 MB and 1 minute long; paid plans up to 200 MB and 15 minutes. The AI stem splitter prepares the file in your browser before uploading it.

Is an AI stem splitter different from a frequency band splitter?

Yes. The AI stem splitter separates vocals, bass and drums. Frequency Band Splitter only filters low, body and high ranges and cannot isolate instruments.

Where does my audio go when I use the AI stem splitter?

To separate a song, your audio is stored privately for a short time while the task runs. Results stay private to your account until you download or save them.

Is this the official Demucs?

No. SampLab Pro is an independent service and is not affiliated with Meta or the Demucs authors. The Local 6 tracks engine runs the open-source HT-Demucs model in your browser. Cloud Standard uses a different third-party model through WaveSpeed. The Licensing page lists both.

Can I paste a YouTube link?

No. SampLab Pro only separates audio files that you upload or pick from your library. It does not download from YouTube or other sites. Use a file that you own or have permission to process.

Can I use the stems commercially?

The rights in a recording stay with its owner, and separating it gives you none. We do not promise that stems are cleared for commercial use, so use songs you own or have permission to use. The Licensing page explains the open points.

What happens to my upload?

Cloud Standard sends the audio to WaveSpeed to make the tracks, and Local 6 tracks never leaves your browser. Uploaded audio and tracks stay private to your account for 7 days and are then deleted automatically. We do not use them to train models, and nobody at SampLab Pro listens to them.