Dataset Open Access
This dataset accompanies the paper:
M.Miron, J.Janer,E.Gomez,"Monaural score-informed source separation for classical music using convolutional neural networks", ISMIR 2017, http://mtg.upf.edu/node/3806
The files are based on Bach10 dataset which comprises 10 Bach chorales: http://music.cs.northwestern.edu/data/Bach10.html
It comprises results in terms of SDR, SIR, SAR as .mat files for the methods presented in the paper.
Additionally, we include audio .wav files for the proposed score-informed source separation method using convolutional neural networks and for the score-informed NMF counterpart.
The code is available at the github repository: https://github.com/MTG/DeepConvSep/tree/master/examples/bach10_scoreinformed
We include the trained CNN model for the proposed approach, which can be used to separate Bach chorales with the code provided at the github repository.