Drop a Song Here
MP3, WAV, FLAC, OGG, M4A — video files work too
How the Vocal Remover Works
Drop a song in and the tool splits it into two tracks: the instrumental and the vocals. It does this with a neural network that reads the song's spectrogram and decides, for every frequency at every moment, how much of that sound belongs to a voice. Nothing is filtered out by frequency the way a karaoke filter would — a vocal in the same range as the guitar is separated from it, which is the whole reason a model is used instead of an equalizer.
The Song Stays on Your Device
The separation model is downloaded to your browser the first time you use the tool — 78 MB, once, then it is kept on your device — and the song itself is never sent anywhere. You can check this in the browser's Network tab: there is no upload. Every other free vocal remover sends your file to a server and processes it there, which is why they can limit you by file size and by the number of tracks per day.
Make a Karaoke Track or an Instrumental
The instrumental track that comes out is a karaoke version of the song: the backing is intact and the lead vocal is gone. It is the usual starting point for singing over a track, for practising an instrument against the original band, or for putting music under a video without a voice competing with the narration. Both tracks come back at full length and in sync with the original, so they line up in any editor.
Where It Struggles, and Why
A separator is not a subtraction, it is an estimate, and it is worth knowing where the estimate gets thin. Heavy reverb on the vocal is partly music by the time it reaches the model and some of it stays in the instrumental. Backing vocals and vocal samples that are part of the arrangement are removed along with the lead, because they are voices. And sibilance — the s and t sounds — lives above the band the model analyses, so a little of it remains in the instrumental. A dense, loud master separates less cleanly than a sparse one.
Keep the Vocals Instead
Both tracks are produced in the same pass, so the acapella is right there beside the instrumental — you can download either or both. If the voice is what you came for rather than the backing, the Vocal Separator page is the same tool with the acapella first and copy written for that job.
Open the Vocal Separator →Split a Song Into Four Stems
Two tracks are enough for karaoke and for most remix work. When you need the drums on their own, or a bass line without the rest of the band, the Stem Splitter runs a four-way model instead: vocals, drums, bass, and everything else. It downloads more and takes longer — 157 MB and around three and a half minutes for a four-minute song — which is why it is a separate page rather than a checkbox here.
Open the Stem Splitter →Then Cut It Down to the Part You Need
An instrumental is usually a means to something else. Send it straight to the Audio Cutter to take the chorus out for a reel, or to the Normalizer to match its loudness to the rest of a project — the result is already in the audio pool, so nothing has to be downloaded and dropped again.
Open the Audio Cutter →Runs in Your Browser
No upload, no queue, no server. The model works on your machine — about a minute for a four-minute song.Nothing Is Uploaded
The song never leaves your device. Verifiable in the browser's Network tab: there is no outbound request.Free, Without an Account
No sign-up, no watermark, no track limit. Both the instrumental and the vocals, at full length.Frequently Asked Questions
Yes. No account, no watermark, no per-track limit. The tool runs on your own device, so there is no server cost per song to pass on — which is exactly why the free online removers that process on a server have to cap you and this one does not.
No. The AI model is downloaded to your browser and the separation happens there. You can confirm it: open the browser's Network tab, run a song through, and you will see the model download the first time and nothing outbound after that.
About a minute for a four-minute song on a normal laptop, plus a one-time 78 MB download of the model on your first visit. After that the model is cached on your device and only the separation time remains. The tool works on one processor thread and does not need a graphics card.
Good enough for karaoke, practice, and most remix work; not good enough to pass as an official instrumental release. Expect a faint trace of the vocal in busy sections, some sibilance left in the instrumental, and reverb tails that stay behind. Sparse arrangements separate more cleanly than dense, heavily compressed masters.
Yes. Both tracks are produced in the same pass and both are offered for download — the instrumental and the acapella. If isolating the voice is your main job, the Vocal Separator page presents the same result the other way round.
MP3, WAV, FLAC, OGG and M4A go in, and video files work too — the audio track is extracted from an MP4 or WebM in your browser. Output is MP3, WAV, FLAC or OGG, picked before you run the separation.
Ten minutes per file. That is lower than the hour the other tools allow, and the reason is memory rather than policy: separation has to hold the whole song in the tab as uncompressed audio plus one copy per track, which is roughly 125 MB per minute. Songs are three to six minutes, so the ceiling is set where the material is.
Spleeter, the source separation model Deezer published under the MIT licence, converted to ONNX and run with onnxruntime in your browser. The licence matters for a tool that is free to use commercially: the stronger research models are published for scientific use only or under non-commercial terms, and cannot honestly be put behind a free public tool.