A new open-source tool called StemDeck lets you pull vocals, drums, bass, and more out of any song, and it all runs on your own machine. According to Hacker News, where the launch climbed to 180 points, StemDeck is a free, local stem separator with no account, no upload, and no subscription. What stands out here is the pitch: your audio never leaves your computer.
You drop in an MP3, WAV, FLAC, OGG/Opus, MP4, or M4A file, or paste a YouTube URL, and the tool splits the track into up to six stems (vocals, drums, bass, guitar, piano, and other). From there you mix it like a producer would.
What StemDeck actually does
- Six-stem separation via Demucs. It runs Meta AI’s open-source htdemucs_6s model and auto-detects the best hardware: CUDA on NVIDIA, MPS on Apple Silicon, and a CPU fallback if you have neither.
- A DAW-style multitrack mixer. Mute, solo, and balance each stem, zoom the waveform, loop a region, and watch live VU meters with peak hold. It behaves like a real mixing console, not a toy web widget.
- Flexible export. Click stem chips to keep only the parts you want, then export individual stems or a single custom mix. Pick a subset and a seventh lane gives you the full song minus your selection, which is handy for A/B reference.
- Song analysis built in. It detects BPM using librosa’s beat tracker, plus key, scale, and confidence, integrated LUFS loudness, and sample peak in dBFS. That’s more musical context than most free tools bother with.
- A real library. Folder-based track organization, drag-and-drop, search, and a trash bin. Jobs are cancellable too, so you can kill a run mid-pipeline and it cleans up the partial files immediately.
How it compares to Moises and LALAL.AI
The maker is refreshingly honest about where StemDeck fits. It’s positioned as a free, local alternative to cloud splitters, and the write-up on Hacker News lays out the trade-offs plainly rather than overselling.
- Price: Free forever, versus freemium plans that need credits or a subscription for regular use.
- Privacy: Your audio stays on your machine. Cloud services upload your files to third-party servers.
- Stem count: Six stems here, versus up to ten on some paid plans.
- Speed: Depends on your hardware. It’s fast with a GPU and slow on CPU only, while cloud tools run on their servers and stay fast regardless of your setup.
- Quality and polish: The commercial products use proprietary models that are generally higher quality, and they ship polished apps. StemDeck calls its own UI “functional, hobby-grade.”
The honest caveats
StemDeck doesn’t pretend to match the commercial crowd. There’s no mobile app, no batch processing (one job at a time), and no extras like pitch shift, chord detection, lyrics, or click tracks. You need an internet connection only for YouTube downloads and the first model fetch, about 170 MB, which is cached afterward for offline use.
The maker is also clear that this is a separation tool, not a downloader. YouTube support exists for content you have the right to process, and StemDeck doesn’t store, cache, or redistribute anything it pulls down.
Why it matters
Stem separation used to be a paid, cloud-locked feature. StemDeck shows how far open models like Demucs have come, and it hands musicians, students, and hobbyists a private, no-strings way to break songs apart for study or practice. It’s significant because it removes both the paywall and the privacy trade-off in one move.
If you need speed, top-tier quality, or mobile access, the commercial products still earn their keep. But if you want stems for personal study and prefer to keep your audio local and free, this covers the core job well. Pre-built installers and zips are attached to each GitHub release, and you’ll find the full feature breakdown at the original source.