Upload a song and get back a vocal track and an instrumental/karaoke track — split apart entirely in your browser, on your own device. No account, no upload to a server, free.
Works best with a normal stereo music file (mp3 or wav) where vocals are panned center — which is how most commercial mixes are made. Mono files can't be processed (see why below).
This tool uses a real, decades-old signal-processing technique called center-channel phase cancellation — sometimes called the "karaoke trick." Most commercially mixed songs pan the lead vocal (and often the kick drum, bass, and snare) dead-center, meaning it's mixed identically into the left and right channels. If you subtract the right channel from the left channel, sample by sample, anything that's identical in both channels cancels out to silence — leaving mostly the instruments that were panned off to one side, like most guitars, keys, and stereo effects. That gives you an instrumental/karaoke track. Do the reverse — add the channels together instead of subtracting — and you keep mostly what was panned center, which is usually dominated by the vocal. All of this math happens on plain audio sample arrays decoded by your browser's own Web Audio API — no file is ever uploaded to a server, and there's no AI model to download, so results appear almost instantly.
This is not the same technology as the real desktop Ultimate Vocal Remover (UVR) application, which uses trained neural network models (MDX-Net, Demucs) to genuinely learn what a vocal sounds like versus an instrument and separate them with studio-grade cleanliness — even when vocals aren't perfectly centered. Phase cancellation is much simpler: it's a geometry trick about stereo panning, not machine learning, and it can't distinguish "vocal" from "any other sound that happens to be centered." That means centered bass, kick, and snare will leak into your "vocal" track, and any vocal doubles, harmonies, or reverb that were panned even slightly off-center will leak into your "instrumental" track. On many pop/rock mixes it works surprisingly well for a quick karaoke backing track; on mixes that are mostly mono, heavily center-panned drums, or already-mono files, it won't do much at all. We'd rather tell you that plainly than oversell it.
Your browser reads the audio file locally and decodes it with the built-in AudioContext.decodeAudioData API — the same decoder your browser already uses to play mp3s and wavs. All of the math (subtracting and adding the left/right channel arrays) runs in plain JavaScript on your device, and the resulting tracks are re-encoded into downloadable .wav files locally, using URL.createObjectURL. At no point does the audio data get sent anywhere as a network request. See the full privacy policy for the technical detail.