Unmixy 4 / 6 stems · 44.1 kHz · runs locally Checking

Unmixany song.

(01)

Vocals, drums, bass, guitar, piano and other instruments on separate tracks. Make karaoke, isolate vocals, separate speech from music in videos, and clean up noisy recordings.

Vocals, drums, bass, other. The cleanest split.
0 MBsent to any server
4 / 6separate stems
$0no sign-up, no limits
—
Preparing
  1. [1] Reading
  2. [2] Model
  3. [3] Splitting
—
00:00.0 / 00:00
Pitch 0 semitones
Speed

(02)How it works

Step 1Drop your file

A song or a video. Nothing is uploaded; it stays on your computer.

Step 2The model splits it

AI runs in your browser and splits the audio into stems. Much faster with a graphics card.

Step 3Mix and download

Turn stems on and off, set their levels, and download karaoke, vocals or your own mix.

(03)Questions

01Is my file uploaded to a server?

No. Splitting happens entirely on your computer, inside your browser. Your file is never sent over the internet.

02Why does something download the first time?

The model that does the splitting is downloaded once on first use (about 87 MB) and stored on your device. It isn't downloaded again.

03How long does it take?

It depends on your computer. With a graphics card a 3-minute song takes about a minute; without one, a few minutes.

04Can I separate speech from music in a video?

Yes. Drop the video, then choose “Speech only” to remove the music or “Remove speech” to remove the speech. Sound effects like doors or cars stay in the background with the music.

05Can I remove noise from a recording?

Yes. Choose “Remove noise” and drop the recording. Hum, fans, wind and room noise are removed and the voice stays. It is made for speech recordings; on music it may treat instruments as noise. You can listen to the removed noise on its own to check.

06Which browsers does it work in?

It is fastest in Chrome and Edge, which can use your graphics card. Other browsers use the processor and are slower.

(04)Technical

ModelHTDemucs v4 and HTDemucs 6s (Meta, open source, MIT license)
Stems4 stems: vocals, drums, bass, other · 6 stems: + guitar, piano (piano is experimental)
Noise removalDeepFilterNet3 (Hendrik Schröter, open source, MIT / Apache 2.0)
Sample rateSplitting 44.1 kHz, noise removal 48 kHz; stereo
OutputWAV, 16-bit
ProcessingIn your browser: WebGPU (graphics card) or WebAssembly (processor)