StemDeck Open-Source Local AI Stem Separator

Quick Take

StemDeck provides free, offline six‑stem separation (vocals, drums, bass, guitar, piano, other) using Meta AI’s open‑source Demucs model, eliminating the need for cloud uploads, accounts, or subscriptions.

What StemDeck Does

  • Local processing only – audio never leaves your machine; YouTube URLs are fetched locally via yt‑dlp.
  • Six‑stem output – uses Demucs htdemucs_6s to isolate vocals, drums, bass, guitar, piano, and an "other" track.
  • DAW‑style mixer – per‑stem volume, mute, solo, monitor, VU meters, waveform zoom, loop regions, and export of individual stems or a custom mix.
  • Cross‑platform installers – native macOS (Apple Silicon & Intel) and Windows builds; also available as a web server or Docker container.
  • No account, no quota – completely free under Apache 2.0, with no telemetry or hidden fees.

Core Technologies

Component Library / Tool Role
Backend Python 3.12 managed by uv Runs inference and orchestrates jobs
API FastAPI REST + Server‑Sent Events for job status
Stem model Demucs htdemucs_6s (Meta AI) Six‑stem neural separation
Optional vocal split audio‑separator (UVR‑MDX‑NET Karaoke 2) Lead/backing vocal extraction
YouTube download yt‑dlp Retrieves audio streams locally
Transcoding / mixing FFmpeg Audio decoding, mixing, export
BPM & key analysis librosa Beat tracking, key/scale detection
Loudness measurement pyloudnorm (ITU‑R BS.1770) Integrated LUFS analysis
Desktop shell Tauri v2 (Rust + WKWebView / WebView2) Packages the UI as a native app
Frontend Vanilla JavaScript + Web Audio API Waveform rendering, playback controls

Feature Highlights

  • Device auto‑selection – picks CUDA on NVIDIA GPUs, MPS on Apple Silicon, or CPU fallback.
  • YouTube & local file import – drag‑and‑drop or paste a URL; supports MP3, WAV, FLAC, OGG/Opus, MP4, M4A.
  • Original backing track – when selecting a subset of stems, a seventh lane shows the complement (full song minus selected stems) for easy A/B comparison.
  • Cancellable jobs – aborts the pipeline instantly and cleans up partial output.
  • Library panel – folder‑based organization, drag‑drop, search, and trash.
  • Export options – individual WAV stems, a summed mix.wav, or a custom MP4 video with the original source.

Honest Comparison with Commercial Services

Aspect StemDeck Typical cloud services (Moises, LALAL.AI)
Price Free, forever Freemium; credits or subscription required
Privacy Audio stays on device Audio uploaded to third‑party servers
Stem count 6 (vocals, drums, bass, guitar, piano, other) Up to 10 depending on plan
Hardware dependence Runs on your GPU/CPU; speed varies Cloud runs on provider’s hardware – fast for any user
Mobile app None iOS/Android apps available
Extra musician tools None (no pitch shift, chord detection, etc.) Often include pitch/key shifting, lyric sync, click tracks
Source code Open‑source, forkable Closed source

If you need maximum quality, batch processing, or mobile access, a paid service may be worth it. For personal practice, privacy, and zero‑cost usage, StemDeck is sufficient.

Getting Started

Desktop Install

  • macOS – download the appropriate DMG (arm64 for Apple Silicon, x64 for Intel). First launch downloads a bundled Python runtime (500 MiB), FFmpeg, and the Demucs model (170 MiB).
  • Windows – unzip the installer (StemDeck-Windows-x64.zip for CPU, *.NVIDIA.zip for CUDA). Run StemDeck.exe; the first launch performs the same runtime setup.
  • Gatekeeper – on macOS, right‑click → Open to bypass the initial security prompt.

Web Server / Docker

  • Python server – git clone https://github.com/stemdeckapp/stemdeck && cd stemdeck && ./run.sh setup && ./run.sh start (macOS/Linux) or equivalent PowerShell commands on Windows.
  • Docker – docker compose -f build/docker-compose.yml up --build. Exposes port 8000; stems appear in ./jobs/.
  • GPU acceleration – set STEMDECK_DEMUCS_DEVICE=cuda and install the CUDA‑enabled PyTorch wheel.

Usage Walkthrough

  1. Import – drag a file onto the bar or paste a YouTube URL.
  2. Select stems – click stem chips; default is all six.
  3. Process – click Process; watch the progress bar through uploading, analyzing, separating, and mixing.
  4. Mix – use the multitrack UI: volume faders, mute/solo, loop region, and playback speed controls.
  5. Export – click Download Mix for a summed WAV, or export individual stems via the file list.

Keyboard shortcuts – Space (play/pause), [ / ] (‑/​+ 5 s seek), L (loop), I/O (loop in/out), M (mute), S (solo), Shift+wheel (coarse volume), wheel (fine volume).

Community Feedback (Hacker News Highlights)

  • Users praise the privacy‑first approach and note that the tool works well for personal practice and learning.
  • Some compare it to Nuo Stems, which uses newer models like mel_band_roformer; StemDeck remains a thin wrapper around htdemucs_6s.
  • A comment points out that Audacity can achieve similar results via OpenVINO plugins, offering an alternative for those who prefer a fully open‑source DAW.
  • Requests for Android support and MIDI conversion of stems appear, indicating interest in extending the workflow beyond desktop.
  • Technical discussion highlights packaging challenges (bundling Python/PyTorch, GPU memory spikes, port collisions) and confirms that StemDeck’s design deliberately avoids large bundled installers by bootstrapping dependencies at first run.

Limitations & Troubleshooting

  • First‑run latency – the Demucs model (~170 MiB) and Python runtime are downloaded on first launch.
  • GPU memory – six‑stem Demucs can exceed 4 GiB VRAM on long tracks; the app falls back to CPU if CUDA is unavailable.
  • ffmpeg not found – install FFmpeg and restart (./run.sh restart).
  • YouTube download warnings – install deno to silence the "No supported JavaScript runtime" message.
  • CPU‑only speed – expect slower processing; a modern GPU dramatically reduces separation time.

License & Disclaimer

StemDeck is released under the Apache 2.0 license. It is intended for personal study, research, and experimentation. The software does not download or redistribute copyrighted material; users must ensure they have the right to process any audio they provide. All third‑party dependencies retain their original licenses.

Where to Find More

Bottom Line

StemDeck delivers a fully offline, open‑source solution for six‑stem audio separation, matching the core functionality of commercial cloud services while preserving privacy and eliminating cost. Its modular architecture, clear licensing, and active community make it a solid foundation for musicians, educators, and developers who need local stem extraction without the overhead of proprietary platforms.

Sources

Related

  • Project
    stemdeckapp/stemdeck

    A free, local audio stem separation tool that uses AI to split songs into isolated tracks like vocals, drums, and bass without requiring cloud uploads.

  • Project
    stemrollerapp/stemroller

    StemRoller is a free application that uses the Demucs algorithm to separate vocals and instrumental stems from songs found via integrated YouTube search.

  • Project
    JeffreyCA/spleeter-web

    A web application for isolating or removing vocals, bass, and drums from songs using AI models like Spleeter and Demucs.

  • Project
    nomadkaraoke/python-audio-separator

    A Python package for audio stem separation that uses UVR models to split audio files into vocals, instrumentals, and other specific instrument tracks.

  • Project
    SUC-DriverOld/MSST-WebUI

    A web-based interface for Music-Source-Separation-Training that allows users to isolate vocals and instruments from music tracks using various AI models.