Vocal remover
Split a song into the music and the voice with an AI model running in this tab. Karaoke tracks and acapellas without uploading anything.
- Nothing is uploaded
- Works offline after first use
- Open source
- No watermark · limit is your device’s memory; phones hold less
Removes the vocals from a song for karaoke, or keeps only the vocals, using an AI separation model (MP3, WAV, M4A, FLAC or video in). It is free, needs no account and adds no watermark, and your files are not uploaded: everything runs in your browser on your device. The first run downloads the separation model (about 67 MB) and the WebAssembly engine that runs it to your browser, and after that the tool works offline.
Done
Open the network tab. Your files never appear in it.
Add files and run the tool to see every request this tab makes.
Your files are processed on this device and never sent to a server. Reload the page to clear everything from memory.
How it works
- Drop one or more songs, or videos with music.
- Choose what to keep: the music without the vocals, the vocals alone, or both as two files, and MP3 or WAV.
- Press Separate. The first run downloads the model and its engine once; each song is split in pieces and the files download when it is done.
In your browser, not on a server
Your browser’s own decoders read the song, from an audio file or a video’s soundtrack, at 44.1 kHz stereo. It is cut into pieces of about six seconds, and each piece becomes a spectrogram (a short-time Fourier transform with a 6144-point window, as the model was trained with). UVR-MDX-NET Inst HQ 3, an MDX-Net model from the Ultimate Vocal Remover project (MIT), predicts the spectrogram of the music without the voice; it runs through onnxruntime-web (MIT) on the graphics chip with WebGPU where the browser has it, or on the CPU with WebAssembly. The answer is turned back into sound, and the vocals are the song minus the music.
Compared with upload-based tools: Online vocal removers upload the whole song, queue it, cap free minutes or songs per day, and ask for an account to download WAV. Here the song never leaves your device, the limit is your device’s memory (phones hold less), and the model works offline after the first run.
Verify it yourself. Give the page a short while after your first run to save itself for offline use, then turn off Wi-Fi or switch to airplane mode and run the tool again. Or open your browser's developer tools (F12, or Cmd-Option-I on a Mac), pick the Network tab, and watch it while you convert. The panel above does the same count for you.
Questions
Is my song uploaded?
No. The song is decoded in this tab and the separation model runs on your device, so unreleased demos, stems for a client and your own recordings stay with you. What gets downloaded is the separation model and the engine that runs it, both from this site, the first time.
How good is the separation?
It uses UVR-MDX-NET Inst HQ 3, one of the models Ultimate Vocal Remover offers for instrumentals, so the music usually comes out clean enough to sing over. Heavy reverb, backing vocals mixed like instruments and very distorted recordings leave faint traces of voice, and the acapella can carry a little of the drums.
Why is it slow on my computer?
The model does a lot of arithmetic. Browsers with WebGPU (recent Chrome and Edge on computers, Safari 26) run it on the graphics chip, which is many times faster. Without WebGPU it runs on the CPU and takes several times as long as the song, with the time left shown as it goes. Keep the tab open until it finishes.
Can it split drums and bass too?
No, this model makes two parts: vocals and everything else.
Can I use the result?
For practice, karaoke at home, remixes of your own songs and study, yes. Separating a track does not change who owns it: publishing an instrumental or acapella of someone else’s song still needs their permission. To sing over the karaoke track, check your microphone first with the mic test.
More video and audio tools
Video background remover
The person stays, the room goes: blurred, a flat colour or a picture of your choice, done frame by frame on your own device.
Audio splitter
Cut one long recording into separate files at the times you choose or at equal lengths. Done on this device, and the file is not uploaded.
Video to MP3
Pull the sound out of a video, or turn any audio file into an MP3. Your files are converted on this device and never uploaded.
Video to GIF
Trim a clip, pick a size and frame rate, and get a looping GIF. The video is decoded by your browser and never leaves this device.
Popular conversions
Guides
- How to remove vocals from a songAI vocal removers pull the voice out of almost any song now, not just tracks with a lucky stereo mix. Here is how they work, and a free browser-based one.