Start with an official acapella if one exists
An official acapella or stem release will always sound cleaner than anything a separation tool produces. Check the artist’s releases, remix contests and DJ promo services first. When there isn’t one, AI stem separation is the practical way to get a usable vocal.
The tools below run on your own computer and are free. We checked each project page on October 6, 2026. Separation gives you a copy of the vocal for your own project; it does not change who owns the recording, so check your rights before publishing anything built on it.
Three free tools that run on your computer
Ultimate Vocal Remover (UVR)
A free, open-source desktop app (MIT licence) with installers for Windows and macOS and instructions for Linux. It runs several separation models, including Demucs, on your own computer. A good first choice if you want a point-and-click app.
Demucs
The open-source separation model behind many vocal removers. It splits a track into vocals, drums, bass and other instruments. It runs from the command line, so it suits people comfortable installing Python tools. Its author notes it is maintained but no longer actively developed.
OpenVINO AI plugins for Audacity
Intel’s free plugins add a Music Separation effect to Audacity that uses a Demucs v4 model and runs locally. Release packages are for Windows; Linux users build from source. Useful if you already edit in Audacity.
Run the separation
Use the highest-quality copy of the song you have. A lossless file or a high-bitrate download separates better than a low-bitrate stream recording, because compression smears the vocal into the instruments.
In UVR, load the file, choose a vocal model (the Demucs and MDX models are common starting points), pick an output folder and process. You get two files: the vocal and the instrumental. Try two or three models on the same chorus and keep whichever leaves the fewest artifacts; different models suit different mixes.
What separation leaves behind
Expect some bleed: hi-hats, reverb tails or a guitar in the same range as the voice can stay in the vocal track. Expect some loss too: breaths and the edges of consonants can sound watery. Dense, heavily compressed mixes are the hardest; sparse mixes with a centred vocal are the easiest.
Listen to the acapella alone and then inside your mashup. Small artifacts that are obvious in solo often disappear once the new instrumental plays underneath.
Clean up the acapella
Cut the silent gaps between phrases rather than leaving low-level bleed running under your new beat. A gentle high-pass filter removes leftover kick and bass. If the separated reverb tail clashes with your instrumental, trim the end of each phrase and add your own reverb instead.
Then measure the vocal: tap its tempo and find its key so you can choose an instrumental that fits. The compatibility checker takes both numbers for each song.
The older trick, and why it is rarely enough
Before AI models, the standard way to remove vocals was to cancel whatever sits in the centre of a stereo mix, which is where lead vocals usually are. Audacity still includes this as its Vocal Reduction and Isolation effect. It takes the bass and kick with the vocal and fails on songs with stereo reverb on the voice, so it is mainly useful for older recordings with simple panning.