%0 Generic %1 subakan2017generative %A Subakan, Cem %A Smaragdis, Paris %D 2017 %K gan source-separation %T Generative Adversarial Source Separation %U http://arxiv.org/abs/1710.10779 %X Generative source separation methods such as non-negative matrix factorization (NMF) or auto-encoders, rely on the assumption of an output probability density. Generative Adversarial Networks (GANs) can learn data distributions without needing a parametric assumption on the output density. We show on a speech source separation experiment that, a multi-layer perceptron trained with a Wasserstein-GAN formulation outperforms NMF, auto-encoders trained with maximum likelihood, and variational auto-encoders in terms of source to distortion ratio.

@misc{subakan2017generative, abstract = {Generative source separation methods such as non-negative matrix factorization (NMF) or auto-encoders, rely on the assumption of an output probability density. Generative Adversarial Networks (GANs) can learn data distributions without needing a parametric assumption on the output density. We show on a speech source separation experiment that, a multi-layer perceptron trained with a Wasserstein-GAN formulation outperforms NMF, auto-encoders trained with maximum likelihood, and variational auto-encoders in terms of source to distortion ratio.}, added-at = {2018-03-20T14:02:39.000+0100}, author = {Subakan, Cem and Smaragdis, Paris}, biburl = {https://www.bibsonomy.org/bibtex/216bf6e40abf41f1d2eb569ca85d6b10f/rcb}, description = {[1710.10779] Generative Adversarial Source Separation}, interhash = {b75ede56ec61d545cf885d218cce2704}, intrahash = {16bf6e40abf41f1d2eb569ca85d6b10f}, keywords = {gan source-separation}, note = {cite arxiv:1710.10779}, timestamp = {2018-03-20T14:03:03.000+0100}, title = {Generative Adversarial Source Separation}, url = {http://arxiv.org/abs/1710.10779}, year = 2017 }