Stemmer uses AI to pull vocals, drums, bass, guitar and piano out of any song or music video. Then you mix them, find the tempo, and export WAV. Your music never leaves your machine.
Free forever · No account · All releases
Features
The same class of AI model as paid online services, running locally, with a real mixer around it.
Vocals, drums, bass and other, or add guitar and piano. Powered by Meta's Hybrid Transformer Demucs.
Volume, mute and solo per stem, waveforms, click-to-seek and keyboard control. Every stem stays perfectly in sync.
BPM and meter detected from the separated drums, so you know what to set your metronome or DAW to.
The AI runs on your computer. No upload, no account, no tracking. Once installed, it works offline.
MP3, WAV, FLAC, M4A and OGG, plus MP4, MOV, MKV and WEBM. Stemmer pulls the audio out of videos for you.
Any single stem, all stems as a ZIP, or your own mix as one file, at full quality.
How it works
No settings to learn. A 5-minute song takes about 2–3 minutes on an Apple Silicon Mac.
Drag in any audio or video file, or pick one. Choose 4 stems, or 6 if you want guitar and piano too.
Demucs separates the instruments on your computer's GPU or CPU, then Stemmer detects the tempo and meter.
Mute the vocals for karaoke, solo the bass to learn a line, then export exactly what you hear.
Why Stemmer
Web services upload your files and charge for more stems or better downloads. Stemmer doesn't.
| Stemmer | Typical online splitter | |
|---|---|---|
| Price | Free, open source | Free tier, then subscription |
| Your files | Stay on your computer | Uploaded to their servers |
| Song limits | None | Daily or monthly caps |
| Guitar and piano stems | Included | Often paid |
| WAV export | Included | Often paid |
| Tempo and time signature | Included | Rarely |
| Works offline | Yes | No |
Download
One download includes everything: the AI models, ffmpeg and the app. No Python or other setup needed.
Apple Silicon (M1 or newer) · macOS 12 or later ·
Windows 10 (1809+) or 11, 64-bit ·
Intel Macs and Linux aren't supported yet. Build from source · Checksums and older versions
FAQ
Yes. Stemmer is open source under the MIT license. There's no paid tier, no account and no limits on songs.
No. The AI model runs on your own computer, and Stemmer only talks to itself on your machine. After installing, it works with no internet connection.
The app isn't signed with a paid Apple Developer ID or Windows code-signing certificate yet, so macOS and Windows show a one-time warning. The source code and build pipeline are public on GitHub, and each release lists SHA-256 checksums.
About 2–3 minutes for a 5-minute song on an Apple Silicon Mac. On Windows the AI runs on the CPU, so expect a few minutes, depending on your processor.
Stemmer uses Meta's Hybrid Transformer Demucs (htdemucs), one of the best open models for music source separation. Vocals, drums and bass come out very clean. Guitar and piano are harder and depend on the mix.
The app is free to use for anything, but the music belongs to its rights holders. Use stems from copyrighted songs for personal practice, study and transcription unless you have permission.