About this website
Contact information
This website is created and maintained by Sevag H (GitHub portfolio, website). If you want to reach out about this website (or any other topic, including brand sponsorships, collaborations, etc.) contact us by e-mail or through our Instagram account. Don’t forget to sign up to the mailing list to receive product updates, promos, free trials, and more!
What is music demixing aka stem separation?
Music demixing (aka stem separation) allows stems for isolated components (vocals, drums, bass, melody) to be extracted from a mixed song. Many songs don’t have original stems available to download publicly. Demixed stems can be used in your own music projects for remastering, remixing, karaoke, instrumental extraction, backing tracks, music education, music analysis, music transcription, and more! The possibilities are endless.
Most demixers consist of complex Artificial Intelligence (AI) models that require a lot of computational power or expensive GPUs to run. Some websites will have a job queue where you submit your track in the cloud and hope it gets processed. This website is easy-to-use and powered by the state-of-the-art Demucs model that runs fast and lean directly in your web browser or Android phone with low memory usage, because we care about speed and efficency! Anybody with a regular computer or phone can use our product.
You can simply choose the track you want demixed and watch as it gets processed immediately on your own device. Your privacy is 100% respected since your files are never uploaded to a server or job queue. Be sure to check the blog for technical articles and deep dives, and read through our privacy policy for more info.
What does it sound like?
Segments extracted from the song Jaxius - Paranoid (Black Sabbath cover) to showcase our models:
Free
Vocals
Drums
Bass
Melody
Pro (deluxe)
Vocals
Drums
Bass
Melody
Segments extracted from the song Periphery - Icarus Lives to showcase our paid models:
Free
Vocals
Drums
Bass
Pro
Vocals (Pro Deluxe, Karaoke)
Drums (Pro Deluxe)
Bass (Pro Deluxe)
PRO subscription
Watch these video tutorials and walkthroughs of the product!
Why pay for this product? What do you get?
All of the demixing on this site is powered by the high-quality Demucs Hybrid Transformer model. ⚠️ By using this site, you acknowledge that your usage of the outputs respects the Meta AI Research Academic/non-commercial licenses ([1], [2])!
Having been involved in music demixing since 2021, I found that:
- Many paid products are secretly using Demucs without telling their users
- Even though Demucs is free to run, it’s not easy to set up, and requires expensive hardware to run
In the current AI world where “bigger is better,” I care about smaller and faster! I bring powerful AI models to your devices, running offline, privately, and with a low memory footprint. This website and the Android app are the easiest way to run the best and latest Demucs-based music demixing models and ensembles available today, all running privately, directly on your device, with unlimited usage!
PRO models
We provide additional models as part of the PRO subscription:
- 6-source: a 6-stem model with vocals, drums, bass, piano, guitar, and other melody
- v3: a super-fast 4-stem model with vocals, drums, bass, and melody
- Karaoke: 2-stem model with specialized high-performance vocals separation
- Fine-tuned: a middle-ground 4-stem model with improved vocals, drums, bass, and melody
- Custom: higher-quality 6-stem model with better piano and guitar separation
- Deluxe: our highest quality model with among the best drums and bass separation available today
Android special features
The Android app has some special features under the PRO subscription:
- Record from the microphone directly in the Music Demixer app
- Capture audio from other apps
- Export a custom mix from the output stems
Technical details
The paid models incorporate the latest research from the 2023 Sound Demixing Challenge to improve the separation quality as measured by the industry standard SDR score (Signal-to-Distortion Ratio):
Execution times
4-minute song, Firefox, 8 workers/32 GB RAM