HomeRitesSermonsThe List
A RITE

THE EXORCIST

Summon the Exorcist once, then purge any track into stems: vocals, drums, bass, guitar, piano and the rest. On your own machine, nothing uploaded to the spirit realm.

OUT OF ORDER — The model currently fails to start in our browser test. Separation and exports are not available reliably. The controls remain here for testing.

OFFER A TRACK

DROP HERE

click or drop · stereo WAV / AIFF / MP3

First run downloads the model (~136 MB) to your browser once, then it can be cached if browser storage is available. Nothing is uploaded. All separation runs on your machine.
HT-Demucs neural separation, run entirely in your browser (WASM). Fast machines finish in a minute or two; long tracks take longer. Empty stems (no piano, no guitar) are hidden.

The Exorcist — stem splitter currently out of order

The Exorcist is a free, browser-based stem splitter currently marked Out of Order after a model-loading failure. It pulls a finished song apart into its separate parts, called stems: vocals, drums, bass, guitar, piano and whatever else is left. Drop a track in and the rite separates it on your own machine, then hands back each stem to preview and download. Nothing is uploaded. The audio never leaves your browser.

Read more. What it does, how to use it, and answers

What it does

Stem separation, also called source separation or demixing, takes a mixed-down song and reconstructs the individual instruments that went into it. The Exorcist runs an HT-Demucs neural model, the same family of model used by most serious stem splitters, entirely inside your browser using WebAssembly. It returns up to six stems: vocals, drums, bass, guitar, piano and a residual "other" track for everything that fits nowhere else. Stems that come back silent, because the song has no piano or no guitar, are hidden rather than handed to you empty.

How to use it

  1. Drop a track in. A stereo WAV, AIFF or MP3 all work.
  2. The first run downloads the model, roughly 136 MB, once. After that it is cached in your browser.
  3. Press Exorcise and wait. The current model-loading failure may prevent this step from completing.
  4. Preview each stem, then download the ones you want.

What it is good for

  • Pulling an acapella or an instrumental out of a finished song
  • Getting a clean drum or bass part to sample, remix or study
  • Making a karaoke or backing track by removing the vocal
  • Learning a part by soloing the instrument that plays it
  • Rescuing a stem when the original session files are long gone

Questions

Is it free?

The Exorcist has no fee or sign-up, but is currently Out of Order.

How do I separate vocals from a song?

Drop the song into the rite and press Exorcise. It returns the vocal on its own, plus the instruments as separate stems, so you can take the acapella or mute the vocal and keep the instrumental.

Is my music uploaded anywhere?

No. The model is downloaded to your browser and the separation runs on your own machine. Your track is not sent to a server. The model can be stored in the browser cache; audio is processed in memory.

Why is the first run slow?

The first run downloads the neural model, about 136 MB. The browser can cache it, subject to available storage and browser eviction. Loading and separation still take time. The separation itself is real computation, so a long track on a slow machine will take a few minutes.

How good are the stems?

Good enough to sample, practise against, remix and build on. No separation is perfect: quiet parts and heavily processed mixes leave artefacts, and a stem the song never really contained will come back thin. Trust your ears.