deezer/spleeter

Spleeter is a tool from the music streaming company Deezer that takes a finished song and pulls it apart into its separate ingredients: the singing voice, the drums, the bass, and so on. This kind of operation is called source separation, because the original recording is one mixed-down sound file and the goal is to recover the individual sources that went into it. Spleeter comes with pretrained models, meaning the heavy work of teaching a computer to recognise voices and instruments has already been done and you just download and use the result. The project offers three flavours of separation: two stems (vocals and everything else), four stems (vocals, drums, bass, other), and five stems (vocals, drums, bass, piano, other). Under the hood it uses Tensorflow, a popular machine-learning framework, and the README says it can produce four-stem separation about a hundred times faster than real time when running on a graphics card. You can use Spleeter from the command line, where one command turns a song file into a folder of split tracks, or call it from your own Python code as a library. It can be installed with pip, used through Docker, or tried out in a hosted Google Colab notebook without installing anything. This is useful if you want to make a karaoke version of a song by removing the vocals, isolate a drum track for sampling, or build any application that needs the individual parts of a recording. The README warns that you should still get permission from the rights holders before processing copyrighted material. The codebase is Python and the project is MIT-licensed. The full README is longer than what was provided.