Skip to content
12notez

Free AI Vocal Remover & Stem Splitter

Free · No sign-up · Files never leave your device

How does an AI vocal remover work?

An AI vocal remover is a neural network trained on songs whose vocals and instruments were recorded on separate tracks. It learns which parts of a finished mix belong to the voice, then splits the song into a vocal stem and an instrumental stem. This page runs two open-source models in your browser: Spleeter by Deezer Research for Fast and HTDemucs by Alexandre Défossez et al. for Studio.

How it works

  1. Step 1

    Your browser reads the song

    The file is decoded by your browser's own audio engine and resampled to 44.1 kHz. Nothing is uploaded.

  2. Step 2

    The model separates it on your device

    A background worker runs the model on your graphics card through WebGPU, or on your processor when WebGPU is not available. It works through the song in 7.8 s segments for Studio and 11.9 s segments for Fast.

  3. Step 3

    You get stems to play and download

    The segments are joined back into full-length stems. Play them together, solo or mute each one, and download any of them as a 16-bit WAV file.

Questions

Is my song uploaded anywhere?

No. Your file is decoded, separated and saved inside this browser tab. The only downloads are the AI model and the engine that runs it, and they come to your device, not from it.

How big is the download?

Fast downloads 25 MB and Studio downloads 158 MB. Each model downloads once and is saved in your browser's storage, so the next song starts without it. You can delete the saved models from this page at any time.

What is the difference between Fast and Studio?

Fast uses Spleeter by Deezer Research and splits a song into vocals and instrumental. It runs in any modern browser, including on phones. Studio uses HTDemucs by Alexandre Défossez et al. and also splits into vocals and instrumental, but at higher quality. It uses WebGPU when your browser has it and your processor when it does not, which is much slower.

Which files work?

Any audio file your browser can play, usually MP3, WAV, M4A, FLAC and OGG. Each stem downloads as a 16-bit stereo WAV at 44.1 kHz.

Why does it offer a shorter clip on my phone?

Separation holds the whole song and every stem in memory at once. Before it starts, the tool estimates how much memory the song needs. If that is more than a phone browser usually allows, it offers a shorter clip instead of letting the tab crash.

Privacy: what runs where

  • Your audio is decoded, separated and turned into WAV files in this browser tab. It is never uploaded.
  • The AI model downloads from Hugging Face (huggingface.co) the first time you use it and is saved in your browser's storage. The engine that runs it (ONNX Runtime) loads from 12notez.com.
  • Nothing about your file is stored on our servers. The saved models stay until you delete them.
  • Only use audio you have the rights to. Stems made from someone else's recording are still their work.

Models: Spleeter by Deezer Research (MIT license); HTDemucs by Alexandre Défossez et al. (MIT license).

Related tools