Free & private · runs on your GPU

Acapella Extractor

Drop a song and the vocal is soloed instantly, with drums, bass, and instruments muted. The AI runs in your browser on your own graphics card, so nothing is uploaded. Full-length acapella files are free in the desktop app.

This browser can't run the separation model yet

The in-browser splitter needs WebGPU (Chrome, Edge, Brave, Arc, or Opera on a computer). Two ways to split stems right now, both free and both on-device:

0:00 / 0:00

Two free splits here. The app does the rest.

Preview downloads here are the first 60 seconds. Xport Studio splits unlimited songs into four full-length WAVs (vocals, drums, bass, instruments) with a choice of separation models, including BS-Roformer. Offline, free forever.

Get the app, free

How acapella extraction works here

The separation model predicts which parts of the mix are the voice and gives you that stem on its own. Doubled vocals, harmonies, and ad-libs are usually captured; heavy reverb tails and vocal chops can partly land in the instruments stem. It runs on your GPU through WebGPU inside the tab; the song never leaves your machine.

What's free here, and what's in the app

Your first two splits on this page are free: mute the vocals for a karaoke pass, solo the vocal to study a take, download a 60-second preview of either stem. This page runs one model (UVR-MDX-NET) and gives two stems. The Xport Studio desktop app is where it gets serious: unlimited songs, four stems (vocals, drums, bass, instruments) or just the acapella or instrumental, full-length WAVs, batch splitting, and a choice of separation models including BS-Roformer, the current state of the art. It runs offline on Mac and Windows and is free forever. We'd rather you use the app for the heavy lifting than make you wait on a browser tab.

Tips for cleaner stems

Lossless or high-bitrate sources (WAV, FLAC, 320 kbps MP3) separate better than low-bitrate files. Heavily compressed masters bleed more between drums and bass. If vocals leak into the instruments stem, that is the model hearing doubled or heavily effected vocals as "other"; the app's higher-quality mode handles those cases better.