wsj0-2mix is a single channel speech separation dataset base on WSJ0. Three speaker extension (wsj0-3mix) is also considered here.
Reference
@article{Hershey_2016,
title={Deep clustering: Discriminative embeddings for segmentation and separation},
ISBN={9781479999880},
url={http://dx.doi.org/10.1109/ICASSP.2016.7471631},
DOI={10.1109/icassp.2016.7471631},
journal={2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
publisher={IEEE},
author={Hershey, John R. and Chen, Zhuo and Le Roux, Jonathan and Watanabe, Shinji},
year={2016},
}