Pretrained models¶
Asteroid provides pretrained models through the Asteroid community in Zenodo. Have a look at the Zenodo page to choose which model you want to use.
Enjoy having pretrained models? Please share your models if you train some,
we made it simple with the asteroid-upload
CLI, check the next sections.
Using them¶
Loading a pretrained model is super simple!
from asteroid.models import ConvTasNet
model = ConvTasNet.from_pretrained('mpariente/ConvTasNet_WHAM!_sepclean')
Use the search page if you want to narrow your search.
You can also load it with Hub
from torch import hub
model = hub.load('mpariente/asteroid', 'conv_tasnet', 'mpariente/ConvTasNet_WHAM!_sepclean')
Model caching¶
When using a from_pretrained
method, the model is downloaded and cached.
The cache directory is either the value in the $ASTEROID_CACHE
environment variable,
or ~/.cache/torch/asteroid
.
Note about licenses¶
All Asteroid’s pretrained models are shared under the Attribution-ShareAlike 3.0 (CC BY-SA 3.0) license. This means that models are released under the same license as the original training data. If any non-commercial data is used during training (wsj0, WHAM’s noises etc..), the models are non-commercial use only. This is indicated in the bottom of the corresponding Zenodo page (ex: here).