//
Split a song into an instrumental and a vocals-only track with an AI model that runs in your browser. The song is never uploaded, and there's no signup, watermark or daily limit.
Your media is processed in your browser and is not uploaded to our servers for conversion. Page assets (including the converter engine) still load over the network.
Select or drop a song
MP3, M4A, WAV, FLAC or a music video. The song is processed on your device.
Click, tap, or drop a fileChoose the song
Drop or select an MP3, M4A, WAV, FLAC or a music video from your device.
Let the model download once
The first time, the AI model (30 MB) downloads and is saved in your browser for next time.
Remove vocals
The song is separated on your device. A typical song takes a few minutes on a laptop.
Listen and download
Play the instrumental and the vocals in the page, then download either or both.
Karaoke and cover singers use the instrumental to sing over. Reels, Shorts and TikTok creators use it as a backing track, and musicians use it to practise a part or learn a song by ear. The vocals-only track (an acapella) is useful for remixes, transcribing lyrics, studying a singer's phrasing, or cleaning speech out of a music bed.
Most vocal removers upload your song to a server, cap you at a few songs a day, or ask for an account before you can download. This one runs on your own device, so it works on private demos and unreleased music too.
The tool uses an MDX-Net model from the open-source Ultimate Vocal Remover project. The song is turned into a spectrogram a few seconds at a time, the model predicts which parts are voice, and those parts are turned back into audio. The instrumental is the original song minus the vocals. Everything runs in your browser with ONNX Runtime, on your graphics card through WebGPU where available, otherwise on the processor.
Results are best on studio recordings with a clear lead vocal. Heavy reverb, backing choirs, live recordings and instruments that sit in the same range as the voice (like a lead violin or saxophone) can leave faint traces. This is a compact model chosen so it can run on ordinary laptops; large server models separate a little more cleanly.
The song stays on your device, so unreleased music and private recordings are safe.
Download the instrumental (karaoke), the vocals (acapella), or both.
No account, no watermark, no songs-per-day cap.
Once the model is saved in your browser, separation doesn't need the internet.
No. Only the AI model is downloaded, once, from Hugging Face. The song is decoded and separated in your browser and never leaves your device.
Roughly 2 to 6 minutes for a 4-minute song on a laptop, depending on the computer. Browsers with WebGPU (recent Chrome and Edge) use the graphics card and are faster. Phones work but take longer. Keep the tab open while it runs.
Clean enough for karaoke and practice on most studio songs. Faint vocal echoes or bits of instruments can remain, especially with heavy reverb, choirs or live recordings.
MP3, M4A, WAV, FLAC, OGG and other audio, or the soundtrack of MP4, MOV, MKV and WebM videos. Songs up to 15 minutes on a computer, or 6 minutes on a phone, because the whole track is processed in memory.
Removing the vocals doesn't change who owns the music. Use instrumentals of songs you have rights to, or follow the platform's rules for covers and karaoke. Personal practice and private karaoke are fine.
Not yet. It splits a song into two tracks: vocals and everything else.
Related: Audio Cutter · Audio Converter · Extract Audio from Video · Transcribe Audio to Text · All tools