The BEST music separation software with help of A.I. ... to my ears ! 👂👂
🥇 Awarded to 1st place by the community as the Best software in the World, for having the less vocal residues in Music track ! Read the Success Story ...
As you've guessed, it was made specially for Karaoke (
Discord community, Chat-GPT & Copilot ... and my human intelligence made this project possible.
KaraFan works on Your PC (Windows 8/10/11, Linux, Mac OSX), if you have a decent GPU. KaraFan works with Google Colab. KaraFan will work with GPU web services API ... (in the future).
You can join all of us on the Dedicated channel on Discord 
(prior to : Discord Invitation on the "Audio Separation" server)
- on Your PC for Users
- on Your PC for Developers
- in Google Colab
- 🔥 Introduction
- 💤 An Old Dream
- 🧒 Biography
- 📆 History
- 🦄 Near Future
- 😍 Wall of Fame
- 💗 Special Thanks To...
- 📜 LICENSE
So for now, in addition to being obliged to R.T.F.R.M (Read This Fucking READ ME), you have to R.T.F.W (Read The Fucking WIKI) !! 😛
~
I'm an amateur Rock singer who has often been disappointed by the inability to find songs of my favorite singers in the vast KAR databases I possess, or in the extensive library of thousands of songs offered by « KaraFun » (with a paid subscription).
Of course, you have all the singers's best-of, but it's not always the songs you want to sing.
And if the singer is not very famous, you'll find 3-4 songs with real musicians playing or you will have to sing on a MIDI file, which is not very pleasant (even with "Sound Fonts"), or you can forget it ... until ...
Since my childhood, I sing everytime and I've been dreaming of a software that would allow me to remove the voice of a song to sing on it.
TODAY, I'M 54 YEARS OLD, I'M STILL SINGING, AND MY DREAM HAS COME TRUE !
Now, I have the ability to create my own « KFN » files for my favorite songs, featuring real musicians who played on the original track, and use them with my beloved software : « KaraFun » 🎤💋
KaraFun is a free software that you can use to play Karaoke files on your PC. They have subscription plans to access their huge library of songs, but you can also use for free with your own MID, KAR, CDG & KFN files !!
Perhaps it's time to turn on your dreams too ... ?
Programmer since the age of 12 (1981). Before, I developed in ASM, C, C++, Basic, Visual Basic. Since the year 2000, I have been coding in PHP, MySQL, JavaScript, jQuery, HTML, CSS.
And today in Python. ❤️
I first started with the Demucs facebook research model, but I was disappointed with the results, especially with the instrumental part.
Then I discovered the MDX model, and I was amazed by the quality of the results, especially with the vocals and the instrumental part.
But I was still disappointed with the instrumental part, which was not shining enough for me.
So I decided to create my own process, based on the MDX models, but with the best instrumental that I could get. (Remembers : It's for Karaoke !).
I also added a few tricks to improve the quality of the results.
~
You will see ... I have hundreds of ideas, but I need time to implement them.
And as I'm an independent developer, I have to work to earn my living.
So if you want to encourage me to give more time to improve this project, you can make a donation or become one of my patrons :
🎵 Go back to the top of this page 🎵 where you can find the donation buttons.
This helps me to :
- Pay calculation time on Google Colab (100 units = 11 €, I eat all in 3-4 days)
- Perhaps, buy new hardware to make more and more tests to improve the quality of the results. I'm currently using a 4 years old laptop with a GTX 1060 Ti 4GB I'm dreaming of a RTX 4090 ... and I'm NOT a gamer ! ... and if you are too generous, Me can buy an Nvidia A100 40GB !! 😍
~
My beloved Patrons & Donators :
~
All these wonderful people without whom this project would not have been possible :
✔️ Jarredou - For his multiples Colab sharing, for his really helpful hand on Discord, and from whom I've stolen ALL the code of inference.py to create my own GUI with it !! 👼
✔️ Deton 24 - For helping hand and for his very BIG documentation, that you can find HERE ! Read it !! ✔️ Bas Curtiz - For his Youtube video : Tips & Tricks - Inverting that confort my FIRST intuition (after thousands tests 😛) about the importance of cascading the models to improve the quality of the results.
✔️ Anjok07 & Aufr33 - for having trained all the MDX-Net AI models that KaraFan uses. ✔️ Roman Solovyev (aka ZFTurbo) - for having trained the last BEST model : « MDX23 - 8K FFT » - (with some assistance from Aufr33 & Anjok). ✔️ Kimberley Jensen - for having trained Kim MDX-Net AI models, especially « Kim Vocal 2 » (that won the 3rd place in the « MDX-C 2022 » challenge)
✔️ Anjok07 & DilanBoskan & Aufr33 - Authors of Ultimate Vocal Remover GUI from where ZFTurbo has stolen some parts of code to create his own... 🙄 ✔️ Roman Solovyev (aka ZFTurbo) - For sharing his code that serve as base to Jarredou & Deton 24 to create their own code. And it's also the creator of MVSEP.com, where you can follow the SDR evolution of the MDX models : Quality Checker !
✔️ Kuielab - Who have developed the original MDX-Net AI code. ✔️ Facebook Research Demucs - For giving me the first steps & the desire to start this project ...
We are all standing on the shoulders of giants !
Thousands Thanks to all of you ! I love you ! ❤️
That means you can do whatever you want with our code, but YOU HAVE TO :
- Mention our names and the fact that WE ARE the original authors of this code
- Keep the original license file in your project
- Copy the header with Copyrights of each source file in the header of your files before redistributing
