Free Vocal Remover
Vocal Remover splits a song into vocals and a karaoke instrumental, or into vocals, drums, bass and other instruments, with Meta's open-source HT-Demucs model. The 126–174 MB model is cached after the first download and runs much faster with WebGPU; Instant mode needs no model. Mix stems with mute, solo and volume, then download WAV, MP3 or a ZIP.
- Stems:
- Vocals + music, or 4 stems
- Max length:
- 12 min (8 with 4 stems, 20 Instant)
- Instant mode:
- No AI, removes the stereo center
- Price:
- Free
- Runs:
- AI model in your browser
- Uploads:
- None: files never leave your device
- Account:
- Not needed
Quick answers
Is my song uploaded?
No. The AI model is downloaded once from Hugging Face and runs in your browser; the audio never leaves your device.
How long does it take?
With WebGPU (recent Chrome or Edge) a 4-minute song takes well under a minute on most computers. Without WebGPU the model runs on the processor and can take 20 minutes or more; the Instant mode takes seconds but is less clean.
Processed on your device: your files are never uploaded. How it works
Network activity in this tab
What this page has requested since you opened the site in this tab. An upload is any request that sends data: POST, PUT, PATCH or a request with a body.
No uploads: nothing has been sent from this page.
No network requests yet.
To see every request, open your browser's developer tools (Network tab).
How to use the Vocal Remover
- Drop a song (MP3, WAV, M4A, FLAC…) or a video with music.
- Choose Vocals + music for karaoke, 4 stems to also split drums, bass and the other instruments, or Instant for a quick result without AI.
- Press the button: the AI model is downloaded once, then the song is separated on your device with a progress bar.
- Listen with the mixer (mute, solo and volume per track) and download each stem, all of them as a ZIP, or the current mix as WAV or MP3.
Credits & open source
Open-source AI model running in your browser. Thank you to the people who make these projects.
- ONNX Runtime WebExecutes ONNX models with WebGPU or WebAssembly (self-hosted from /vendor/ort). © Microsoft Corporation · MIT
- MediabunnyReads, writes and converts media files with WebCodecs (MP4, WebM, MKV, MOV, MP3, WAV, Ogg...). © Vanilagy and contributors · MPL-2.0
- lamejsMP3 encoder (port of LAME), loaded as a separate unmodified chunk. © LAME project; JavaScript port by Alex Zhukov; ESM fork by Breezystack · LGPL-3.0
- fflateFast ZIP compression for batch downloads. © Arjun Barrett · MIT
Web platform APIs used: Web Workers, WebAssembly, WebGPU, Web Audio API, File API.
- HT-Demucs v4 for WebGPU (ONNX, mixed precision, 21 pieces) by Simon Rouard, Francisco Massa, Alexandre Défossez (Meta AI, weights); ONNX export by monteslu (MIT) — AI stem separation on WebGPU with half-precision support
- HT-Demucs v4 (ONNX, fp32) by Simon Rouard, Francisco Massa, Alexandre Défossez (Meta AI, weights); ONNX export by Ghilda (MIT) — AI stem separation on the CPU and on WebGPU without half precision
Full license texts: third-party notices · all credits
Frequently asked questions
Why do I still hear a bit of the voice?
Separation is an estimate: reverb, backing vocals and instruments in the same range as the voice can leak between tracks. The AI modes are much cleaner than the Instant mode, which only removes what sits in the center of the stereo image.
What is the difference between the modes?
Vocals + music gives a vocal track and a karaoke track. 4 stems also separates drums, bass and the remaining instruments. Instant needs no download and works best on studio songs with a centered lead vocal.
Can I use the result?
The tool and the AI model are free and open, but the song itself keeps its copyright: only use stems of music you have the rights to, or for private practice.