Add robustness to automatic decoder instantiation

kensho-technologies / pyctcdecode

A fast and lightweight python-based CTC beam search decoder for speech recognition.

Apache License 2.0

419 stars 89 forks source link

Closed gkucsko closed 2 years ago

gkucsko commented 3 years ago

This diff adds better parsing for alphabets as well as sanity checks + logging around potential issues during instantiation

sarim-zafar commented 3 years ago

any update on this?

gkucsko commented 3 years ago

Yeah sorry, was moving states, but most of the code it already done, will update and hopefully land code tomorrow

jedzill4 commented 2 years ago

Is this still in development ? I found some of the same issues/errors fixed on this branch and i was waiting for this merge to look in deep

gkucsko commented 2 years ago

my bad sorry, I got busy with other projects. Will get back to this very soon

gkucsko commented 2 years ago

@jedzill4 @sarim-zafar sorry for the delay, the new code should now be up on pypi as v0.1.1. Please let us know if there are still issues.

jedzill4 commented 2 years ago

Thanks for the work! it works smoothly