How the Stem Separator works
Why it takes a minute or two — and why that's a good thing.
Your song never leaves your device
Most vocal removers upload your song to their servers, separate it there and send the result back. The Stem Separator does the whole job inside your web browser instead: the song is read, split and saved on your own computer or phone. Nothing is uploaded, nothing is stored anywhere else, and there are no accounts or limits.
The trade-off is that your device does the heavy lifting itself, so it takes a little longer than a big server would.
What's actually happening
Separating a song is done by an artificial intelligence model called HTDemucs (Hybrid Transformer Demucs), made by Meta AI and released as open source. It's one of the best models there is for this job — the same kind of technology many paid apps use.
The model has been trained on thousands of songs to recognise what vocals, drums, bass and other instruments sound like. It works through your song a few seconds at a time and, for every moment, works out which part of the sound belongs to which instrument. That takes billions of calculations per song, which is why you see a progress bar.
When it's done you get four stems — separate tracks for the vocals, the drums, the bass and everything else (guitars, keys, synths and so on). Played together, they add back up to the original song.
The first time
The first time you use it, the model itself (about 190 MB) is downloaded to your browser. It's then kept on your device, so the next songs start straight away. If you clear your browser's data, it's simply downloaded again.
How long it takes
- Computers with a recent browser (Chrome, Edge, Firefox or Safari) usually use the graphics chip, and a typical song takes about a minute or two.
- Without a usable graphics chip it runs on the processor, using all of its cores — a few minutes for a song.
- Phones and tablets work too, but take longer and need plenty of free memory.
The progress bar shows an estimate of the time left as it goes.
Tips to make it faster
- Keep the Stem Separator tab open and in front while it works — browsers slow down tabs in the background.
- Plug in your laptop: on battery, many computers slow themselves down.
- Close other heavy tabs and apps, especially games and video editors.
- Use an up-to-date browser. Chrome and Edge are usually the fastest.
- Long songs take longer: up to 15 minutes of audio can be separated at once.
How good is the result?
For most pop, rock, hip-hop and electronic music the stems are very clean. Very dense mixes, live recordings and heavily processed vocals can leave faint traces of one instrument in another stem — that's normal for every separator, including paid ones.
What you can do with the stems
- Make a karaoke / instrumental version (everything except the vocals).
- Get the vocals alone (a cappella) for remixes or covers.
- Practise along with a song with your own instrument muted.
- Download any stems separately in a ZIP, or mixed together into one file — as MP3, WAV, FLAC, M4A or OGG.