Stem splitting and audio conversion running in-browser with Rust/WASM

关于

I got tired of all the ugly "free" audio conversion tools littered with ads so I built my own that runs entirely on your machine in the browser. It does stem separation (vocals/drums/bass/other), format conversion, BPM detection and speed/pitch changes. Nothing is uploaded (there is no api) + the models runs client-side. Stack is Rust compiled to WASM, mediabunny for conversion + WebGPU. Getting a separation model to run at usable speed in the browser was the most fun part of this project. Excited to play around with other models as they get small enough to load client-side! Stem separation quality is behind the best server-side models (ElevenMusic) but still relatively impressive. A 4-minute track takes roughly 20s on an M-series Mac and much longer on integrated GPUs. Memory pressure on long files and Safari's WebGPU support are still rough edges but conversion and BPM are solid. Hope others find value in these audio tools, let me know if you'd like to add anything. I plan on open-sourcing the code.

收录记录

1. Show HN

   2026年10月1日 · 1 次收录