Spleeter: Extract voice, piano, drums, etc. from any music track (github.com)

Article summary

Spleeter is an open-source music source separation library developed by Deezer, which can extract vocals, piano, drums, and other instruments from any music track. The library uses pre-trained models and can be used straight from the command line or as a Python library. Spleeter is fast and can perform separation of audio files 100x faster than real-time when run on a GPU. It has been used by professional audio software and has a commercial version called Spleeter Pro.

Main themes

• music source separation
• audio processing
• machine learning
• music analysis
• copyright and ownership
• audio software
• music creation

What commenters say

• Using Spleeter to separate audio tracks can be a useful tool for music analysis and creation.
• The availability of large datasets of multitrack recordings, such as those from Rock Band games, could be used to train and improve music separation models.
• Some commenters have successfully used Spleeter to separate audio tracks and have shared their results online.
• The use of Spleeter raises questions about the ownership and accessibility of separated audio tracks, particularly for copyrighted music.
• The quality of the separated tracks can be impressive, with some commenters noting that the reverb in the vocal track is preserved.
• There are potential applications of Spleeter beyond music, such as separating unrelated pieces of audio.
• The use of chiptune music, such as from the SNES, could provide a viable source of training data for music separation models.