How to split a song into stems
Drop the song on the page, choose 2 or 4 stems and press Split. Listen with mute and solo, then save the stems as WAV files.
Add a song
Drop an MP3, WAV, FLAC, M4A or OGG file on the page, or a video: its sound is used. Your browser opens it right here.
Choose 2 or 4 stems
2 stems gives vocals and an instrumental. 4 stems gives vocals, drums, bass and everything else (guitars, keys, strings). The first time, the AI model downloads to your browser; after that it is ready at once.
Split, listen and save
The split runs on your computer's processor. When it is done, play the stems together, mute or solo any of them and set their levels, then save one stem or all of them in a ZIP.
2 stems or 4 stems?
| 2 stems | 4 stems | |
|---|---|---|
| You get | Vocals, instrumental | Vocals, drums, bass, other |
| Good for | Karaoke and backing tracks, a cappellas | Remixes, drum or bass practice |
| First download | 54 MB | 93 MB |
Both use Deezer's Spleeter models, trained on thousands of songs. 2 stems runs about twice as fast, and its instrumental is cleaner than adding up drums, bass and other from 4 stems.
Where does the song go?
It stays on your device. The AI model downloads from this site once, and your browser keeps it. Your song is split in the background and the stems are saved to your Downloads folder.
- Decoded by your browser's own audio decoder
- Split by an AI model running on your processor (WebAssembly)
- Saved to your device
Specs and limits
| Input | MP3, WAV, FLAC, M4A (AAC, ALAC), OGG, Opus, AIFF, and the sound of MP4, MOV and WebM videos: whatever your browser can decode |
|---|---|
| Length | Up to 20 minutes per song |
| Output | WAV, 16-bit, 44.1 kHz, stereo (mono for a mono song). One file per stem, or all in one ZIP |
| AI model | Spleeter by Deezer (MIT licence), converted for the browser and checked against the original. Sees up to 11 kHz; the stems above it follow each stem's share just below, so cymbals and air are kept and the stems add up to the song |
| Speed | Depends on your processor: on a recent laptop, a 4-minute song takes roughly half a minute with 2 stems and a minute with 4 |
| Browsers | Current Chrome, Edge, Firefox and Safari on a computer. Phones work for short songs but are slow |
| Price | Free, no account, no watermark |
Questions
No. The AI model is downloaded to your browser once and runs there. The song is decoded, split and saved on your device.
2 stems for a karaoke track or an a cappella: vocals and everything else. 4 stems to remix or to practise an instrument: vocals, drums, bass and other. 2 stems downloads 54 MB once and runs faster; 4 stems downloads 93 MB once.
The model estimates which part of each frequency belongs to which stem, and where voice and instruments share a frequency, a trace of one stays in the other. Reverb on the voice is the most common leftover. Spleeter is fast and small enough to run in a browser; the largest server models separate more cleanly.
The tool puts no limits or watermark on them. The rights to the song stay with its owners: splitting a song does not change who owns it, so ask before you publish anything made from someone else's recording.