[Whisper] Add labels' in the whisper output

[wu6u3tw]

The original text output from whisper-large-v3 includes numbers and the normalization part are included in the accuracy_eval script.
Therefore, to get the digit's part of the output in the label dict. I add digits, some symbols in the labels.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[wu6u3tw]

recheck

[keithachorn-intel]

I'm not sure this will align with the text normalization elsewhere in the reference. Numerical values were previously expanded to full words. Will have to check how the OpenAI normalizer handles numeric values.

[hanyunfan]

I'm not sure this will align with the text normalization elsewhere in the reference. Numerical values were previously expanded to full words. Will have to check how the OpenAI normalizer handles numeric values.

@wu6u3tw Could you review Keith’s question and provide additional context or a possible response?

[keithachorn-intel]

Per TF meeting, let's remove the euro, lb, and cent symbols since they are interacting strangely with VIM editor and don't appear to impact measured accuracy. Also, let's replicate this change in reference_SUT.py file.

[keithachorn-intel]

Per TF meeting, let's remove the euro, lb, and cent symbols since they are interacting strangely with VIM editor and don't appear to impact measured accuracy. Also, let's replicate this change in reference_SUT.py file.

[wu6u3tw]

Remove euro, cent, lb sign done.

[keithachorn-intel]

Re-reviewed and approved. Thanks!

[keithachorn-intel]

Reviewed and approved by Speech-to-text TF.

[keithachorn-intel]

@pgmpablo157321 - This PR is approved per the TF discussion today. I think we can close it before the next WG sync (no need for wider input).

[hanyunfan]

LGTM