Local stem separation (Vocal/Instrumental)

[basharovV]

By now most people have heard of AI stem separation tools (extract vocals, instruments). I wanted to include this in the app, but use only local, open source models. Thought it would be useful eg. karaoke, musicians looking to study.

I've got a prototype working, details below. Is this something people are interested in? Happy to hear thoughts, model suggestions, etc..

This is currently on the main branch only, but I expect to publish a release in the next few days.

How it works

• It uses portable-vocal-remover which is a Rust port of Ultimate Vocal Remover.
• PVR is embedded into Musicat as a sidecar binary, allowing the app to spawn it to split stems. This is also better for performance (not sharing thread poll with Tauri app) and app size (the PVR binary is only 2.5 MB)
• Right-click track -> "Separate stems":

• PVR runs in the background, reporting progress, and when completed it deposits the resulting stems to your "Generated stems location" folder (by default it's /Users//Music/Musicat/Stems:

Kapture.2025-09-27.at.16.54.33.webm

Dependencies

• To avoid bloating the app, stem separation needs the following files in your config directory - the ONNX runtime for your system, and the ONNX model itself:

You need to manually download and place these files on your system (eventually these will be downloaded automatically).
You can download the ONNX runtime binaries from here: https://github.com/microsoft/onnxruntime/releases/tag/v1.23.0
You can download the UVR model from here: https://huggingface.co/Blane187/all_public_uvr_models

[JombaGo]

Some ideas for you:

1. allow showing the artworks of each song in the library.

2. Feature to display local lyrics (embedded and .lrc)(unsynced and synced(word-by-word as well)

3. Ability to search artist images, both automatic and manual selection)