Desktop · audio DSP · polyglot systems
A local Windows desktop app that turns audio you own into a vocal-free producer sound kit — stems, one-shots, MIDI, and analysis — fully offline. Ships as native MSI/NSIS installers.
Started June 2026 · self-directed
The problem
Producers want stems, loops, and MIDI out of reference audio they have the rights to — but cloud tools mean uploading copyrighted audio, and they don't run offline.
How it works — three processes, one app
The story · STAR
Situation
As a producer I wanted stems, loops, and MIDI out of reference audio I own, but cloud tools mean uploading copyrighted audio, and I wanted everything offline.
Task
Build a local Windows desktop app that turns authorized audio into a usable sound kit, end to end.
Action
Built the UI in Tauri + React, but Python did the DSP — and getting them to talk reliably on Windows was the hard part, so I put a small Node HTTP bridge between them. The worker separates stems, analyzes, slices, and generates MIDI, then exports a ZIP. I kept job state in SQLite, abstracted separation behind a provider interface, added a Playwright smoke test, and packaged it into native MSI/NSIS installers.
Result
A feature-complete app with working Windows installers — the full upload-to-export pipeline runs locally and was verified on 2026-06-19. The lesson was architectural: when one runtime fights the platform, a thin bridge process can be cleaner than forcing it.
Download (Windows 10/11 · x64)
Two installer formats. The owner publishes these on a GitHub Release — verify the SHA-256 after downloading.
MSI installer · 2.82 MB · click to download
NSIS installer · 1.87 MB · click to download
Verify, then install:
Heads up: the installer isn't code-signed, so Windows SmartScreen may warn ("Windows protected your PC") — click More info → Run anyway if the checksum matches.
Built with
Status & honest limitations
Feature-complete & build-verified (2026-06-19), with shipped MSI/NSIS installers.
dsp-lite separation works offline; high-quality ML separation is an opt-in provider slot (user-approved download).Self-directed project, built with AI-assisted development. Local-first: your audio is never uploaded.