Open
Description
🚀 The feature
Hi 👋
Hope you are having a good day, I was thinking to use CTC Beam Search on my Handwritten Recongizer (parseq) model.
Question#1: Is there something already implemented??
Question#2: If I want to use something like, https://github.com/githubharald/CTCWordBeamSearch, What would be the easiest way to get output tensor from recognizer model?
Motivation, pitch
Since handwriting is a difficult area to handle and my model is only 1 or 2 character off from the original word, I though it would be good if we can get help from a corpus or language model to fix those mistakes.
Alternatives
No response
Additional context
No response