Spaces:

adalat-ai

/

wer-analysis

like 0

Sleeping

For many applications, it may be useful to ignore the punctuation and/or case when evaluating word error rate. This PR adds checkboxes to ignore punctuation and case, and then applies the relevant transforms to the ground truth and hypothesis text.

I contemplated using jiwer's off the shelf transformations as defined here - https://jitsi.github.io/jiwer/reference/transformations/, which can directly be passed into the process_words function. However, this didn't have the option to remove punctuation.

Additionally, I also add typing to the python file and format the file.