Zenodo.org will be unavailable for 2 hours on September 29th from 06:00-08:00 UTC. See announcement.

Dataset Open Access

Bach10 Separation SMC2017

Marius Miron

The Bach10 Separation SMC2017 dataset is derived from the Bach10 dataset, which contains ten pieces of Bach chorales along the scores.
We separate the audio files in the original dataset and in the dataset we synthesized with Sibelius (https://zenodo.org/record/321361#.WLW40t-i7J8), using the approaches presented in this paper:
Marius Miron, Jordi Janer, Emilia Gomez, "Generating data to train convolutional neural networks for low latency classical music source separation", Sound and Music Computing Conference 2017

The dataset contains the separated audio files along the computed measures which give the quality of separation: SDR, SIR, SAR, computed with BSS Eval 3.0. 

For the intellectual rights and the distribution policy of the original dataset check the Bach10 dataset page:

The files in Bach10 Separation SMC2017 dataset are offered free of charge for non-commercial use only. You can not redistribute them nor modify them. 

This dataset is created by Marius Miron, Music Technology Group - Universitat Pompeu Fabra (Barcelona). This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 Unported License.

Files (1.1 GB)
Name Size
Bach10 Separation SMC2017.tar.gz
1.1 GB Download
All versions This version
Views 503504
Downloads 7171
Data volume 79.0 GB79.0 GB
Unique views 450451
Unique downloads 6464


Cite as