Separate vocals. No upload.
Your recording stays on your device. Separate vocals and music in this browser, then save both WAV files. Free use needs no payment, account, installation or daily cloud allowance.
Free: up to 10 minutes / 100 MB per file, with no daily local quota. First use downloads about 100 MB; cached models can be reused. Mobile compatibility is not guaranteed.
Choose audio to separate
Start with a 30–60 second excerpt. Accepts mono or stereo audio up to 10 minutes and 100 MB. Use a recording you have the right to process.
Separate on this device stays unavailable until the local engine is ready. Prepare the engine on this screen first.
Off by default. Sends only action counts while online, never audio, filenames or settings. Site-wide analytics consent applies separately. You can switch it off at any time. Privacy
Prepare your local engine
First use downloads about 100 MB of model and runtime files. The model can be reused from this browser’s cache; the browser may remove them later.
Free: 10 minutes / 100 MB per file. No daily local quota.
Previously purchased access can still be restored with your checkout email or checked in this browser. Free local use has the same file limits.
· Sign in to restore purchaseExperimental: speed and separation quality depend on your device and recording. CPU mode can take much longer; slower devices may need shorter files. Keep this tab open while processing. A failure stops locally and never switches to an upload.
Uses UVR-MDX-NET-Voc_FT by Ultimate Vocal Remover and its developers Anjok07 and aufr33, under the MIT statement in the official UVR README. MDX-Net architecture credit: Kuielab & Woosung Choi. Vocal bleed and separation artifacts are possible. The model is downloaded from a checksum-verified Hugging Face mirror of the official GitHub release; that host receives model requests, never your audio. Models are cached when possible, while recordings and results stay only in this tab. Model sources and licenses.
Local separation is free within the stated file limits and uses no cloud minutes. The $9.99 / 90-minute / 12-month cloud Minute Pack is separate. Training-data rights have not been verified; the maintainers have not provided written confirmation for this individual model file. You must have permission to process your source audio.
Back to the main separator · The main tool uses uploaded audio. Choosing this link does not upload or transfer your selected recording.