[AUDITORY] Announcing Asteroid: PyTorch-based audio source separation toolkit for researchers

[Manuel Pariente <pariente.mnl@xxxxxxxxx>]

=== Apologies for cross-posting ===

Dear list,

We are happy to present Asteroid, the PyTorch-based audio source separation toolkit for researchers. Asteroid provides code to train state-of-the-art deep learning models on common audio source separation and speech enhancement datasets.

Asteroid's most noticeable features:

• Asteroid comes with architectures such as Deep Clustering, Chimera++, TasNet, ConvTasNet, DualPathRNN, DualPathTransformer, DCUNet, DCCRN.
• Permutation invariant training is made extremely simple with our PITLossWrapper.
  
• Asteroid supports several filterbanks, losses, metrics and datasets for easy experimentations.
• Our model hub hosted on Zenodo enables us to share and download pretrained models.
• All the code is released under the MIT License.
  

Webpage : https://asteroid-team.github.io/
Code : https://github.com/mpariente/asteroid
Docs : https://mpariente.github.io/asteroid/
Model hub : https://zenodo.org/communities/asteroid-models
Paper : https://arxiv.org/abs/2005.04132
Presentation video : https://www.youtube.com/watch?v=imnZxQwuNcg

We hope that this resource will be useful to the community, and we invite any one of you to help us grow it in the directions you would like to see.

All the best,
Manuel Pariente