Vocal Separation code @Email
Vocal Separation code @Email
主题
Dear all,
I want to separate vocals from single channel music recordings that contain singer voice and musical accompaniment. I am now using the RPCA, and need to find a few algorithms to compare the effect of vocal separation. Do you have algorithms that have been implemented.
Best regards and thank you!
Matt McVicar，
In our paper (http://www.mattmcvicar.com/wp-content/uploads/2016/04/mcvicar.pdf) we made use of REPET-SIM (http://www.zafarrafii.com/repet.html) and this system (http://paris.cs.illinois.edu/pubs/huang-ismir2014.pdf, https://github.com/posenhuang/deeplearningsourceseparation). Code is available for both of these systems.
Bernhard，
there is also a Matlab implementation of the KAM (Kernel Additive Modeling) Algorithm:
http://www.loria.fr/~aliutkus/kam/
Antoine，
Thanks Bernhard for linking to KAM !
Maybe I can add that we released KAML http://www.loria.fr/~aliutkus/kaml/ that fixes the RAM issues of KAM. The preset is for music/voice separation, but please be aware that it works better on longer excerpts, where the musical structure can repeat several times.
You may also want to check FASST, that implements NMF-like separation http://bass-db.gforge.inria.fr/fasst/
as well as the VIMM/VUIMM 'source filter models', another variant of NMF http://www.durrieu.ch/research/jstsp2010.html
I am not aware of a publicly available off-the-shelf method for vocals separation based on DNN, does someone know any ? We will soon be releasing one here at India, and will keep you posted
Justin,
In case you find it useful for your comparative evaluation, mir_eval implements a number of standard source separation evaluation metrics and is pretty straight forward to use.
Carl,
Fun! I'm working on a LSTM-RNN single-channel source separation method for extracting vocals in the exact setting you describe and I'm just now finishing up a Python package that does this. It will be published in two weeks. I'll let you know then if you'd like to test it.
