Separate audio tracks into isolated instrument stems locally

StemDeck splits a song into up to six stems, vocals, drums, bass, guitar, piano and other, and opens them in a studio where you can mute, solo, slow down, loop, change the key and play along with a click. Drop in an audio file or paste a YouTube or SoundCloud link. Separation runs on your own machine, and StemDeck does not upload your audio.

A separation tool, not a downloader. StemDeck is for processing audio you have the right to use. What you import, and the stems made from it, are saved in your library folder on the machine that runs StemDeck; StemDeck does not upload or share them. See the Disclaimer.

Download

Each link downloads the latest release. Release notes · Every version

Platform	NVIDIA or Apple Silicon	Everything else
Windows	NVIDIA GPU	No NVIDIA GPU
macOS	Apple Silicon (M1 and later)	Intel Mac
Linux	NVIDIA GPU	No NVIDIA GPU

 Docker / Unraid: ghcr.io/stemdeckapp/stemdeck:latest. It runs on the CPU as it is; add --runtime=nvidia to use an NVIDIA GPU. On Unraid, search "StemDeck" in Apps.

• Guide. A new button in the left rail turns the studio into its own manual: point at anything to see what it does, and every number in the song facts says what it measures.
• The Trash. A count on its icon, hold a song for a second to bring it back, and pick several at once with Ctrl or Shift.
• Deleting a song while it extracts stops it, and keeps its files until you empty the Trash.
• Imports stay in the background: the song you have open stays open.
• We Recommend welcomes Adrianna Claro, and cards can now link to Spotify and Apple Music.
• Portable Windows installs in a folder they cannot write to now say so clearly.

All of it, with credits to the people who reported each problem, is in the release notes.

Windows and Linux. Unzip, then run StemDeck.exe (Windows) or StemDeck (Linux). No installer and nothing else to install. Put the folder somewhere you can write to, such as C:\StemDeck: a portable StemDeck keeps its runtime, models and settings in a data folder beside it, so it cannot run from Program Files. The first launch downloads FFmpeg and the separation model, and on an NVIDIA card sets up GPU support; later launches start in seconds.

macOS. Open the DMG and drag StemDeck to Applications. StemDeck is not code-signed yet, so clear the quarantine flag once, or macOS will say the app is damaged:

xattr -dr com.apple.quarantine /Applications/StemDeck.app

The first launch downloads the runtime and the model. Apple Silicon separates on its built-in GPU.

Docker and Unraid. See Build from source for the docker run line, GPU passthrough and the Unraid volume mapping.

Updates. The desktop app tells you when a new version is out and installs it for you.

What you can do

Split any song

• Up to six stems with Meta's open Demucs model: vocals, drums, bass, guitar, piano and other.
• Lead and backing vocals as separate lanes, on demand (not on Intel Macs).
• Files or links: MP3, WAV, FLAC, OGG, Opus, M4A and MP4, or a YouTube or SoundCloud link.
• Search without leaving the app: YouTube songs and playlists, and SoundCloud, with a preview.
• Pick only what you need before you extract.
• Whole YouTube playlists go into an import queue and run one after another.

Practice with it

• Mute or solo any part, and set each one's volume.
• Slow it down to half speed, or anywhere from 0.50x to 0.99x, without changing the pitch.
• Loop the hard part, by dragging across the waveform.
• Change the key, of the whole song or of one part.
• Play with a click that follows the song, with a count-in, any beats per bar and groupings such as 3+2+2.
• Fix the beat grid when the click drifts: move, add or remove beats and mark the bar lines.

Know the song

• Key, tempo and loudness, found for you.
• Sections such as verse, chorus and solo, detected for you (experimental) or marked by hand.
• Synced lyrics from LRCLIB, with timing you can fix by tapping along. With an NVIDIA GPU, StemDeck can also write them down from the vocals when none are found.
• Song and artist details from MusicBrainz, Wikipedia and, optionally, Discogs.

Keep and share

• Export what you hear as one file, every part as a .zip, only the looped part, or, when the song came with a video, a video with the original picture. Add the click or a count-in if you like.
• Drag a stem or a loop from the desktop app straight into your other music software.
• A library with colored folders, tags you can search by (#tag), favorites, and a Trash that only deletes when you empty it.

It explains itself

Click Guide in the left rail and point at anything. Every button, lane control and number says what it does, in plain words. While Guide is on, clicks explain instead of acting, so nothing happens by accident.

On your phone too

Turn on Make StemDeck available on your network in Settings and scan the QR code. Your phone opens your library and mixer over your own Wi-Fi, with speed and key controls. No app store, no cloud.

Honest comparison

StemDeck is not trying to beat the commercial stem-separation services. It covers the core well and stops there. This table is here so you can choose knowingly rather than find the gaps later.

If you need the best quality, speed on any machine, or a store-bought phone app, the commercial products are worth the money. If you want stems for practice and study, keep your audio private, and like software with no strings attached, StemDeck is enough.

We Recommend

StemDeck is free and does not accept any money, sponsorship, or funding from anyone listed below. I share these makers and artists and communities purely for the joy of pointing you toward wonderful people doing beautiful work. Go meet them ❤️

Star history

For developers

StemDeck is built on Python 3.12 managed via uv, with a FastAPI backend serving REST and Server-Sent Events. Stem separation uses Demucs (htdemucs_6s), Meta AI's open-source 6-stem neural network. The optional on-demand lead/backing vocal split runs the UVR-MDX-NET Karaoke 2 model via audio-separator, trained as part of the Ultimate Vocal Remover project by Anjok07. YouTube audio is fetched via yt-dlp; transcoding and mixing use FFmpeg. BPM detection and key analysis run on librosa; loudness measurement uses pyloudnorm (ITU-R BS.1770). The macOS and Windows desktop shells are Tauri v2 (Rust/WKWebView on macOS, Rust/WebView2 on Windows). The frontend is vanilla JS with the Web Audio API, no framework and no build step; waveforms are rendered on <canvas> using min/max sample rendering.

Thanks to the creators and maintainers of all the open-source libraries that make StemDeck possible.

Disclaimer

StemDeck is a local audio stem separation tool intended for personal study, research, and experimentation. It is not a downloading service. Audio you import, and the stems made from it, are saved only in the library folder on the machine that runs StemDeck, and StemDeck does not upload, share or redistribute them. To find song details and lyrics it sends the song's name to the services named in its Settings and, only if you add an AcoustID key, an audio fingerprint; it does not send the audio itself.

YouTube and SoundCloud link support is provided via yt-dlp as a convenience. Automated downloading may violate YouTube's Terms of Service. You, the user, are solely responsible for ensuring you have the right to process any audio you submit, complying with the terms of service of any site you download from, and respecting the copyright of the material you work with.

You are also responsible for following the licenses of the underlying tools this project depends on (yt-dlp, Demucs, FFmpeg, PyTorch, and others listed in pyproject.toml).

The author(s) of StemDeck provide this software "as is", without warranty of any kind, and accept no responsibility or liability for how it is used.

License

StemDeck is Apache-2.0.

Every download ships a THIRD_PARTY_NOTICES.txt and a licenses/ folder. licenses/INDEX.txt lists each packaged Python dependency with its version and license, and each one's full license text sits beside it. That inventory is generated from the packaged interpreter at build time, so it describes what actually shipped rather than what was expected to.

FFmpeg is a GPL build. StemDeck downloads it and runs it as a separate executable, so it does not change StemDeck's own license, and THIRD_PARTY_NOTICES.txt carries the written offer of source for the exact build your platform receives.

Community

Questions, ideas and show-and-tell are all welcome.

Contributing

Issues, feature suggestions and pull requests are welcome. See the open issues for what is planned, and Build from source to run it yourself.