Update TTS pronunciation dictionaries

[quapham]

What changed

• Add scripts/tts_dataset_files/hi_IN/hi_prondict-v0.3_nostress_fixed.dict.
• Update scripts/tts_dataset_files/ipa_cmudict-0.7b_nv26.07.txt A/A(1) pronunciation ordering.

Scope

Dictionary-only change. No code changes.

Validation

• Verified copied dictionary file hashes match the source files.
• Verified git diff contains only the two dictionary files.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[blisc]

Can you bump the version of scripts/tts_dataset_files/ipa_cmudict-0.7b_nv23.01.txt to nv26.07? Can update references from 23.01 to 26.07?

[quapham]

Updated the IPA CMU dictionary filename/references from nv23.01 to nv26.07 and pushed commit ca999d9

[quapham]

/ok to test ca999d9

[quapham]

/ok to test ca999d9

[quapham]

/ok to test 4dfe3a5

[XuesongYang]

what happened to this dictionary?

[quapham]

We trained the model with the updated dictionary, where the pronunciation was changed from /'eɪ/ to /'ʌ/.

[XuesongYang]

/ok to test 4dfe3a5

[quapham]

/ok test 8ebea33

[blisc]

/ok to test 8ebea33

[XuesongYang]

@quapham pls make bugfix for the failed unit test, where

E           FileNotFoundError: [Errno 2] No such file or directory: 'scripts/tts_dataset_files/ipa_cmudict-0.7b_nv23.01.txt'

[XuesongYang]

@blisc nit, our doc has two broken links to be fixed. Do you have any thoughts?

The failure is from Sphinx linkcheck (docs build with --fail-on-warning) and is caused by 2 broken URLs not in exclusions:

asr/speech_classification/datasets.rst:126
Broken: https://cs.taltech.ee/staff/tanel.alumae/data/voxlingua107/ (404)

tts/checkpoints.rst:1
Broken: https://ngc.nvidia.com/catalog/models/nvidia:nemo:tts_en_mixertts (404)

[quapham]

/ok test 5a31913

[quapham]

I already updated the external TestData YAML through the Internal-Developer S3 copy config. However, this unit test can still load the old nv23.01 hparams

I added a compatibility migration for text_tokenizers..g2p.phoneme_dict to rewrite ipa_cmudict-0.7b_nv23.01.txt to ipa_cmudict-0.7b_nv26.07.txt.