SonArcan

Dive into the music.

A focused, local-first desktop workspace to learn, analyze, isolate, and rehearse music.

Note

SonArcan Full targets Apple Silicon, NVIDIA GPUs on Windows/Linux, and AMD GPUs on Linux. SonArcan Light targets all supported platforms without bundling the heavy analysis models or runtimes.

Minimum and recommended configuration

Platform	Minimum for playback, lyrics, spectrum and meters	Required for Beat, Chords and Mix	Recommended
macOS Apple Silicon Full	macOS 14, M1, 8 GB RAM	A working, qualified MLX/MPS accelerator	M2 or newer, 16 GB RAM
macOS Apple Silicon Light	macOS 14, M1, 8 GB RAM	Not included	16 GB RAM for large projects
macOS Intel Light	macOS 12, Intel x64, 8 GB RAM	Not included	macOS 13+, 16 GB RAM
Windows NVIDIA GPU	Windows 10 1903 or newer, x64, 16 GB RAM	NVIDIA GPU and current driver compatible with CUDA 12.6; startup model probe must pass	Windows 11, 8 GB GPU memory, 32 GB RAM
Windows Light	Windows 10 1903 or newer, x64, 8 GB RAM	Not included	Windows 11, 16 GB RAM
Linux NVIDIA GPU	Ubuntu 22.04-compatible x64 desktop, glibc 2.35+, 16 GB RAM	NVIDIA GPU and current driver compatible with CUDA 12.6; startup model probe must pass	Ubuntu 22.04 or newer, 8 GB GPU memory, 32 GB RAM
Linux AMD GPU	Ubuntu 22.04-compatible x64 desktop, glibc 2.35+, 16 GB RAM	AMD GPU and driver supported by ROCm 7.2; startup model probe must pass	Ubuntu 22.04.5 or newer, 8 GB GPU memory, 32 GB RAM
Linux Light	Ubuntu 22.04-compatible x64 desktop, glibc 2.35+, 8 GB RAM	Not included	Ubuntu 22.04 or newer, 16 GB RAM

SonArcan Full checks the production accelerator and model graphs once when the application starts. If no compatible and qualified GPU backend is available, it enters a safe degraded mode for the complete session: Beat, Chords, Mix, BPM, and the analysis-driven metronome are not shown and cannot be started. Playback, time navigation, lyrics, spectrum, and the stereo meter remain available. The explanatory message is shown only once per user profile.

SonArcan Light is a deliberately smaller edition. It keeps playback, imports, projects, time navigation, loops, training, pitch/tempo controls, lyrics, spectrum, and the stereo meter, but does not include Beat, Chords, Mix, BPM, or the analysis metronome. Full and Light use the same .sac project format and never delete analysis data produced by another edition.

Windows and Linux GPU bundles contain a pinned accelerator-specific PyTorch runtime: CUDA 12.6 for NVIDIA, and ROCm 7.2 for AMD on Linux. PyTorch exposes ROCm through its CUDA-compatible API, but the packages and installer remain separate. Windows AMD and Intel GPUs are not qualified in this beta; choose Light on those systems. SonArcan never silently falls back to the CPU for heavy analysis jobs.

GPU runtimes exceed GitHub's 2 GiB limit for a single release asset. Their portable package is consequently published as numbered part-000, part-001, … files plus a SHA256SUMS file. Download every part for the chosen platform and backend into one directory, verify the checksums, then concatenate them in name order. On Linux this reconstructs an installable .deb:

cd ~/Downloads
version=v0.1.0-beta.24
backend=NVIDIA # Replace with AMD for the ROCm release.
sha256sum --check "SHA256SUMS-Linux-${backend}-GPU.txt"
cat "SonArcan-Linux-x86_64-${backend}-GPU-${version}.deb".part-* > "SonArcan-${backend}-GPU.deb"
sudo apt install "./SonArcan-${backend}-GPU.deb"

Replace NVIDIA with AMD for the ROCm build. On Windows, verify the hashes with Get-FileHash, concatenate the numbered files as binary data, then extract the reconstructed .zip and launch SonArcan NVIDIA GPU.exe:

$ErrorActionPreference = 'Stop'
Set-Location "$HOME\Downloads"
$version = 'v0.1.0-beta.24'
$checksumFile = 'SHA256SUMS-Windows-NVIDIA-GPU.txt'
foreach ($line in Get-Content -LiteralPath $checksumFile) {
  $expected, $file = $line -split '\s+', 2
  $actual = (Get-FileHash -LiteralPath $file -Algorithm SHA256).Hash.ToLowerInvariant()
  if ($actual -ne $expected.ToLowerInvariant()) { throw "Checksum mismatch: $file" }
}
$parts = @(Get-ChildItem "SonArcan-Windows-x86_64-NVIDIA-GPU-$version.zip.part-*" | Sort-Object Name)
if ($parts.Count -eq 0) { throw 'No archive parts found' }
$archive = "SonArcan-NVIDIA-GPU-$version.zip"
$output = [IO.File]::Create($archive)
try {
  foreach ($part in $parts) {
    $input = $part.OpenRead()
    try { $input.CopyTo($output) } finally { $input.Dispose() }
  }
} finally { $output.Dispose() }
Expand-Archive -LiteralPath $archive -DestinationPath "SonArcan-NVIDIA-GPU-$version"
& ".\SonArcan-NVIDIA-GPU-$version\SonArcan NVIDIA GPU.exe"

Light releases and both macOS releases remain ordinary one-file installers.

Choose an edition

Feature	Full / GPU	Light or degraded mode
Playback, pitch/tempo, loops and trainer	Yes	Yes
Lyrics	Yes	Yes; moves into the Mix column
Spectrum and stereo meters	Yes	Yes
Beat timeline, BPM and analysis metronome	Yes	No
Chord detection and navigation	Yes	No
Piano, guitar and ukulele chord views	Yes	No; assets are excluded
Six-stem Mix and export	Yes	No

Download SonArcan for an Apple-silicon Mac, SonArcan NVIDIA GPU for a supported NVIDIA Windows/Linux computer, SonArcan AMD GPU for supported AMD Linux hardware, or SonArcan Light everywhere else. Detailed notes for the current beta are in RELEASE_NOTES.md.

SonArcan is made for musicians who want the useful parts of an audio workstation without the weight of a full DAW. Import a setlist, understand the music, build loops, and practice while keeping projects portable and data on your computer. Full editions also analyze the music and isolate its parts locally.

✨ Highlights

Import WAV, MP3, FLAC, local files, or YouTube sources into portable .sac projects.
Play, seek, change gain, and create seamless A/B loops through a dedicated Rust audio engine.
Slow down or speed up from 50–200% independently of pitch, with ±12 semitones and fine cent correction.
In Full editions, detect BPM, beats, downbeats, and timed chords locally, with detected timelines and source-aware disposable caches.
In Full editions, separate six stems locally with HTDemucs 6s through MLX or portable Torch, then mix or export them.
Practice with a progressive loop trainer, waveform, spectrum, and stereo meter; Full editions add the synchronized analysis metronome.
Keep per-track practice settings, recent projects, diagnostics, and a multilingual interface.

For planned work and known product directions, see the roadmap.

🧰 Built with—and grateful for

SonArcan stands on an outstanding open-source audio and desktop ecosystem:

Area	Tools and projects
Desktop & interface	Rust, Tauri 2, Svelte 5, TypeScript, Vite
Real-time audio	CPAL, Symphonia, Signalsmith Stretch, RustFFT
Source separation	Apple MLX, demucs-mlx, PyTorch, HTDemucs 6s, Python
Musical analysis	LV-Chordia, Beat This!, PyTorch, librosa
Import & media	FFmpeg, LAME, yt-dlp
Reproducible builds	npm, Cargo, uv, GitHub Actions

A heartfelt thank-you to every maintainer, researcher, tester, and contributor behind these projects. Their work makes SonArcan possible. Licensing and attribution details are collected in Third-party notices.

🚀 Run from source

Requirements

macOS 14+ on Apple Silicon, macOS 12+ on Intel for Light, Windows x64, or a Linux x64 desktop supported by Tauri 2
Node.js 22+ and npm
Stable Rust 1.78+ with Cargo
uv exactly 0.9.26
FFmpeg and FFprobe on PATH for development fallback
Tauri 2 prerequisites for the target OS

Install the native tools for your operating system first. The FFmpeg command below is macOS-specific; use your distribution package manager on Linux or put FFmpeg and FFprobe on PATH on Windows.

rustup toolchain install stable
brew install ffmpeg
curl -LsSf https://astral.sh/uv/0.9.26/install.sh | sh
uv python install 3.13.5

On Windows PowerShell, install the same pinned uv release with:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/0.9.26/install.ps1 | iex"
uv python install 3.13.5

Every profile starts with a fresh checkout and the frontend dependencies:

npm ci

Choose exactly one of the following profiles.

Apple Silicon Full: MLX + MPS

This is the only source profile that prepares the shared HTDemucs model. MLX handles six-stem separation; PyTorch MPS handles Beat and Chords.

npm run mlx:sync
npm run mlx:model
npm run chords:downbeat-model
npm run python:runtime
npm run ytdlp:search
npm run ffmpeg:runtime
npm run quality
npm run tauri dev -- --config src-tauri/tauri.macos-arm.conf.json

Windows or Linux Full: Torch GPU

Run npm run stems:sync, not mlx:sync. These profiles require the verified config.json and htdemucs_6s.safetensors produced by the Apple Silicon prepare-model release job under src-tauri/resources/models/demucs-mlx/. Copying an arbitrary model into that directory will fail its identity checks.

For NVIDIA on Linux or in a Unix-like Windows shell:

export SONARCAN_EDITION=full
export SONARCAN_GPU_BACKEND=nvidia
npm run stems:sync
npm run chords:downbeat-model
npm run python:runtime
npm run verify:gpu-runtime
npm run ytdlp:search
npm run ffmpeg:runtime
npm run quality
npm run tauri dev -- --config src-tauri/tauri.nvidia-gpu.conf.json

For AMD ROCm on Linux, replace nvidia with amd and use src-tauri/tauri.amd-gpu.conf.json. Windows AMD and Intel GPU Full profiles are not qualified. Native Windows PowerShell sets the NVIDIA environment with:

$env:SONARCAN_EDITION = 'full'
$env:SONARCAN_GPU_BACKEND = 'nvidia'

Then run the same npm commands without the two export lines.

Light

Light does not prepare MLX, Torch, Beat This!, LV-Chordia, or HTDemucs. It keeps playback, projects, imports, lyrics, spectrum, meters, and time-based practice.

export SONARCAN_EDITION=light
npm run ytdlp:search
npm run python:light-runtime
npm run verify:light-runtime
npm run ffmpeg:runtime
npm run verify:ffmpeg-release
npm run quality
npm run tauri dev -- --config src-tauri/tauri.portable.conf.json

Use src-tauri/tauri.macos-arm-light.conf.json or src-tauri/tauri.macos-intel-light.conf.json instead on macOS. In PowerShell, set the edition with $env:SONARCAN_EDITION = 'light' and omit the export line.

npm run dev starts only the frontend. Playback, project management, native menus, analysis, and stems require npm run tauri dev and the Rust backend.

📦 Desktop bundles

The contents of a desktop bundle depend on its edition. End users never need to install Python, uv, FFmpeg, or model dependencies themselves.

Build profile	Targets	Analysis implementation	Bundled resources
MLX Full	Apple Silicon	`sonarcan-mlx-worker` for six-stem separation and PyTorch MPS for Beat/Chords	MLX/MPS runtime, HTDemucs, Beat This!, LV-Chordia, FFmpeg and yt-dlp
Torch GPU Full	Windows/Linux NVIDIA; Linux AMD	`sonarcan-torch-worker` using CUDA 12.6 or ROCm 7.2 for six-stem separation and Beat/Chords	Backend-specific PyTorch runtime, HTDemucs, Beat This!, LV-Chordia, FFmpeg and yt-dlp
Light	Apple Silicon, Intel Mac, Windows x64 and Linux x64	No ML worker and no heavy analysis	Minimal Python runtime for yt-dlp plus FFmpeg; no Torch, MLX or analysis models

The tag workflow is the authoritative cross-platform build recipe: it chooses the correct worker, accelerator runtime, edition environment, resources, and Tauri configuration for each target. It verifies the packaged resources before leaving the release as a draft for manual smoke testing.

GitHub macOS builds are ad-hoc signed so every embedded executable has a consistent code signature, but they are not notarized or identified by Apple. On first launch, users must explicitly allow SonArcan under System Settings → Privacy & Security → Open Anyway. The complete workflow and trust model are documented in the release guide.

🎼 Portable projects

A .sac project is an inspectable directory. macOS presents it as a single SonArcan document package; Windows and Linux keep the same portable contents:

My-Band.sac/
├── project.json
├── Audio/
├── Stems/
├── Analysis/
├── Chords/
├── Lyrics/
└── Cache/

The manifest stays human-readable. Original media and user-authored data are kept separate from disposable analysis and cache files.

📚 Documentation

Architecture · Development · Quality
Real-time audio · Chord analysis · Stem separation
Practice workflow · Project management · Waveforms
Competitive analysis · Roadmap · Release guide
Contributing · Security

🤝 Contributors

SonArcan grows through code, ideas, testing, musical feedback, and careful open-source work. Thank you—warmly—to everyone who has helped shape it.

Current repository contributors:

OrangeJuce82 — creator and maintainer

Want to join the list? Read CONTRIBUTING.md, open an issue, or submit a focused pull request. Every thoughtful contribution is welcome.

☕ Support SonArcan

If SonArcan helps your practice sessions and you would like to support its continued development, you can offer the project a coffee through PayPal:

Thank you for listening, testing, contributing, sharing, and supporting the project. 💙

License

SonArcan is available under the MIT License. Third-party components and models retain their own licenses; see THIRD_PARTY_NOTICES.md.

Name		Name	Last commit message	Last commit date
.github/workflows		.github/workflows
.idea		.idea
docs		docs
scripts		scripts
src-tauri		src-tauri
src		src
tools		tools
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CAHIER_DES_CHARGES.md		CAHIER_DES_CHARGES.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
RELEASE_NOTES.md		RELEASE_NOTES.md
SECURITY.md		SECURITY.md
THIRD_PARTY_NOTICES.md		THIRD_PARTY_NOTICES.md
index.html		index.html
osv-scanner.toml		osv-scanner.toml
package-lock.json		package-lock.json
package.json		package.json
svelte.config.js		svelte.config.js
tsconfig.json		tsconfig.json
vite.config.ts		vite.config.ts

Navigation Menu

Repository files navigation

SonArcan

Minimum and recommended configuration

Choose an edition

✨ Highlights

🧰 Built with—and grateful for

🚀 Run from source

Requirements

Apple Silicon Full: MLX + MPS

Windows or Linux Full: Torch GPU

Light

📦 Desktop bundles

🎼 Portable projects

📚 Documentation

🤝 Contributors

☕ Support SonArcan

License

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

SonArcan

Minimum and recommended configuration

Choose an edition

✨ Highlights

🧰 Built with—and grateful for

🚀 Run from source

Requirements

Apple Silicon Full: MLX + MPS

Windows or Linux Full: Torch GPU

Light

📦 Desktop bundles

🎼 Portable projects

📚 Documentation

🤝 Contributors

☕ Support SonArcan

License

About

Resources

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages