'Automatic pretrained weight download' feature will help user to use this library easier.
I test this functionality through torch.hub.
Can I make a pull request?
Current Status
Users can install the x-vector model class through pip, but needs manual weight download (e.g. wget, git clone).
This step needs engineering skills to ML users, so brings non-streamlined developer experience.
This situation could prevent more widespread usage of this valuable package.
Cause
Manual weight download is the cause of this pain.
Proposal
We could use torch.hub system for automatic weight (and package) download.
With this system, we can instantiate the model without any explicit model/package download as below:
model = torch.hub.load("tarepan/xvector_jtubespeech-official:feature/hub", "xvector", trust_repo=True)
torch.hub.load automatically:
pull the xvector_jtubespeech package from github
download the xvector.pth file from github
Access _XVector class and instantiate it
load state_dict
I wrote basic feature of torch.hub in my fork.
If you kindly consider reviewing this feature, I am grad to make a pull request.
Notes
When I tested ,this x-vector model is so good at Japanese speaker classification.
Thanks for your great model & package.
Summary
'Automatic pretrained weight download' feature will help user to use this library easier.
I test this functionality through
torch.hub
.Can I make a pull request?
Current Status
Users can install the x-vector model class through pip, but needs manual weight download (e.g.
wget
,git clone
).This step needs engineering skills to ML users, so brings non-streamlined developer experience.
This situation could prevent more widespread usage of this valuable package.
Cause
Manual weight download is the cause of this pain.
Proposal
We could use
torch.hub
system for automatic weight (and package) download.With this system, we can instantiate the model without any explicit model/package download as below:
torch.hub.load
automatically:xvector_jtubespeech
package from githubxvector.pth
file from github_XVector
class and instantiate itI wrote basic feature of
torch.hub
in my fork.If you kindly consider reviewing this feature, I am grad to make a pull request.
Notes
When I tested ,this x-vector model is so good at Japanese speaker classification.
Thanks for your great model & package.