SonArcan v0.1.0-beta.25

SonArcan 0.1.0-beta.25

This beta introduces platform-specific Full GPU and Light releases. Choose the
installer whose name matches your computer; all editions share the same .sac
project format.

Which file should I download?

Release name	Computer	Beat, Chords, Mix	Included compute runtime
SonArcan	Apple-silicon Mac (M1 or newer)	Yes	Apple MLX and MPS
SonArcan NVIDIA GPU	Windows x64 or Linux x64 with a compatible NVIDIA GPU	Yes, after the startup probe succeeds	PyTorch CUDA 12.6
SonArcan AMD GPU	Linux x64 with a ROCm 7.2-compatible AMD GPU	Yes, after the startup probe succeeds	PyTorch ROCm 7.2
SonArcan Light	Apple-silicon Mac, Intel Mac, Windows x64, or Linux x64	No	No ML runtime or models

There is no AMD GPU edition for Windows in this beta. AMD's Windows support is
currently limited to selected recent GPUs and requires a separate Python 3.12
runtime, which has not yet completed SonArcan's release qualification. Use
SonArcan Light on an AMD-only Windows computer. Intel GPUs are not qualified yet.

Full GPU releases never run Beat, Chords, or Mix silently on the CPU. At every
application launch, SonArcan exercises the actual production model graphs on the detected
accelerator. If the driver, device, runtime, model, memory, or inference result
is incompatible, SonArcan enters safe degraded mode for that session. Beat,
Chords, Mix, BPM, the analysis metronome, and the piano/guitar/ukulele chord
views are hidden; playback, time navigation, lyrics, spectrum, and stereo meters
remain available. The explanation is shown once per user profile.

Light is the smallest and safest download for older hardware. It physically
excludes Torch, MLX, the analysis models, and the chord-instrument frontend
assets rather than merely hiding them.

GPU download format

CUDA and ROCm runtimes are too large for GitHub's 2 GiB limit per release file.
Each GPU package is therefore portable and split into numbered part-000,
part-001, … files, accompanied by a platform/backend-specific SHA256SUMS
file. Download every part for one edition and its matching SHA256SUMS file
into the same directory. Light and macOS downloads remain conventional
single-file installers.

Linux NVIDIA or AMD

Open a terminal in the download directory, set backend to NVIDIA or AMD,
then run:

cd ~/Downloads
version=v0.1.0-beta.25
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"

Do not install the reconstructed package if sha256sum reports a missing file
or a checksum failure.

Windows NVIDIA

Open PowerShell in the download directory and run:

$ErrorActionPreference = 'Stop'
Set-Location "$HOME\Downloads"
$version = 'v0.1.0-beta.25'
$checksumFile = 'SHA256SUMS-Windows-NVIDIA-GPU.txt'
foreach ($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"

PowerShell stops before reconstruction if a part is missing or altered.

Other fixes

• Waveform clicks now seek to the exact pointed position without beat, chord,
  or lyric snapping; keyboard and transport navigation remain unchanged.
• Startup now loads and applies user preferences before rendering the workspace,
  then keeps a localized bootstrap screen visible until the recent or temporary
  project is active.
• Windows desktop builds no longer leave a console window open.
• Intel macOS FFmpeg assembly now falls back safely when NASM is unavailable.
• Full and Light editions use distinct product names and bundle identifiers.

See the README for detailed minimum configurations and installation guidance.

SonArcan v0.1.0-beta.18 — Lyrics in time with your music

SonArcan beta.18 brings lyrics directly into the practice workspace. Import, find, edit, follow, navigate, and export lyrics without leaving the track — while keeping your music and project data local.

What’s new

• Import lyrics in the formats musicians actually use: plain text, LRC, enhanced LRC with word-level timing, and Apple-style TTML.
• Follow playback live: the current line — and timed words when available — stays highlighted as the song plays.
• Find lyrics without an account: optional LRCLIB search ranks matches against the recording duration and prefers synchronized results.
• Keep practicing offline: once selected, lyrics are stored inside the portable SonArcan project; playback never depends on the provider.
• Edit in place: adjust text or timing, apply a display offset, and turn provider results into a local project copy.
• Navigate by lyrics: synchronized line starts join Time, Beat, and Chord as a navigation mode. Use the transport controls or Left/Right arrows; press N to cycle through the modes currently available.
• Export your work: save synchronized lyrics as LRC — including enhanced word timing — or export a clean Markdown document from the native Songs menu.

Built for safe, focused practice

Lyrics documents are validated and bounded before they are saved. Malformed timestamps and timings outside the audio duration are rejected, writes are atomic, and the original audio is never modified. Provider credentials are not required and network requests are not stored in the project.

This release also tightens repository hygiene so generated models, media tools, and platform runtimes stay out of Git history while remaining reproducible through the release pipeline.

Downloads

Choose the installer for your platform:

• macOS 14+ / Apple Silicon: SonArcan_0.1.0-beta.18_aarch64.dmg
• Windows x64: SonArcan_0.1.0-beta.18_x64-setup.exe
• Linux x64: SonArcan_0.1.0-beta.18_amd64.deb

The application bundles its audio tools, Python runtime, analysis models, and six-stem separation backend. No separate Python, FFmpeg, or model installation is required. The downloads are large because these local-first components are included.

First launch on macOS

The macOS build is ad-hoc signed, but it is not notarized or identified by Apple. After trying to open SonArcan for the first time:

1. Open System Settings → Privacy & Security.
2. Scroll to Security.
3. Click Open Anyway next to SonArcan.
4. Confirm with Touch ID or your Mac password.

macOS remembers this choice for later launches.

Integrity

The SHA-256 checksum for the macOS DMG is:

12b0fc00e2755f2cfe7e4daaa412280d86bbf8cc5ecb20d762a4cbba08283f14

A matching SHA256SUMS.txt file is attached to the release.

Beta note

This is a pre-release intended for real-world testing. Reports about lyric matching, timing, format compatibility, accessibility, and cross-platform packaging are especially welcome through GitHub Issues.

Full changelog: v0.1.0-beta.17...v0.1.0-beta.18

Dive into the music. 🎵