Drop a Song Here
MP3, WAV, FLAC, OGG, M4A — video files work too
How to Separate Vocals From Music
Drop a song in and press Separate. A neural network reads the song's spectrogram and decides, for every frequency at every moment, how much of that sound is a voice — then hands you the voice as one track and the music as another. Both come back at the original length and in sync with the source, so they drop straight back into an editor against the original.
What an Isolated Vocal Is Good For
An extracted acapella is the starting point for a remix, a mashup, or a sample. It is also how people study a performance: with the band gone, phrasing, timing and breath are audible in a way they never are inside the mix. Producers use it to lift a hook, transcribers use it to hear a lyric that the master buries, and singers use it to compare their own take against the original line for line.
The Song Never Leaves Your Browser
The model downloads once — 78 MB, then it stays on your device — and the song itself is never sent anywhere. That is checkable rather than a claim: open the browser's Network tab and run a track through. Tools that separate on a server have to meter you by file size and by tracks per day, because every song costs them compute; this one costs us nothing per song, which is why there is no limit on it.
How Clean the Acapella Is
Clean enough to sample and to study, not clean enough to pass as a studio stem. Expect some bleed from instruments that share the voice's range — a lead synth or a distorted guitar in the same octave is the hardest case — and a slightly dulled top end, because the model analyses frequencies up to about 11 kHz and the air above that is left with the music. Reverb and delay on the original vocal are partly separated and partly left behind, since a reverb tail is as much room as it is voice.
Or Keep the Music Instead
Both tracks come out of the same pass, so the instrumental is already there beside the acapella and you can download either. If the backing track is what you actually want, the Vocal Remover page is this tool with the instrumental first and its copy written for karaoke and practice.
Open the Vocal Remover →Drums and Bass on Their Own
This page splits a song two ways. When a remix needs the drum loop without the bass line, or the bass without the keys, the Stem Splitter runs a four-way model: vocals, drums, bass, and everything else. It is a bigger download and a longer wait — 157 MB and roughly three and a half minutes for a four-minute song — so it lives on its own page rather than as an option here.
Open the Stem Splitter →Acapella and Music
Both tracks from one pass, full length, in sync with the original. Download either or both.Nothing Is Uploaded
The song stays on your device. Verifiable in the browser's Network tab: no outbound request.No Account, No Limit
No sign-up, no watermark, no cap on tracks per day. The work happens on your machine.Frequently Asked Questions
Drop the file in and press Separate Vocals. The tool splits the song into a vocal track and a music track and offers both for download. Nothing is uploaded — the model runs in your browser — and there is no account or limit.
Good enough to sample, to transcribe, and to study a performance. Not a studio stem: instruments sharing the voice's range bleed in a little, the very top end is slightly dulled, and reverb tails are partly left with the music. Sparse mixes separate better than dense, loud masters.
Same model, opposite output. A vocal remover keeps the instrumental and throws the voice away; this page keeps the voice. Both tracks are produced either way, so whichever page you start on, the other track is right there.
Not on this page — it splits two ways. The Stem Splitter uses a four-way model and returns vocals, drums, bass and everything else. It downloads 157 MB instead of 78 and takes about three and a half minutes for a four-minute song.
Ten minutes per file. Lower than the hour the other tools allow, and the reason is memory rather than policy: separation holds the whole song in the tab as uncompressed audio plus a copy per track, roughly 125 MB per minute.
Spleeter, released by Deezer under the MIT licence, exported to ONNX and run with onnxruntime in the browser. The licence is the reason for the choice: the research models that separate better are published for scientific use only or under non-commercial terms, which no free public tool can honestly build on.