Add Arabic char tokenizer and Japanese-English katakana support by quapham · Pull Request #15614 · NVIDIA-NeMo/NeMo

quapham · 2026-04-16T06:08:42Z

Important

The Update branch button must only be pressed in very rare occassions.
An outdated branch is never blocking the merge of a PR.
Please reach out to the automation team before pressing that button.

What does this PR do ?

Add Arabic character tokenizer support and improve mixed Japanese-English katakana handling in Japanese G2P/tokenizer tests.

Collection: [Note which collection this PR will affect] TTS, common

Changelog

add Arabic character tokenizer coverage/support
update Japanese katakana accent handling for mixed English/Japanese text
add unit tests for tokenizer behavior

Usage

You can potentially add a usage example below

# Add a code snippet demonstrating how to use this

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

Before your PR is "Ready for review"

Pre checks:

Make sure you read and followed Contributor guidelines
Did you write any new necessary tests?
Did you add or update any necessary documentation?
Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
- Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

New Feature
Bugfix
Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

Related to # (issue)

XuesongYang · 2026-04-20T07:22:16Z

thanks for adding new tokenizers.

Regarding 4 arabic locales, we decided to use separate tokenizers for ar-SY, ar-AE, ar-MSA, and ar-SA. In this PR, I only found ArabicCharsTokenizer was defined. Could you pls add tokenizers below?

  ar-AE: [arabic_AE_chartokenizer]
  ar-MSA: [arabic_MSA_chartokenizer]
  ar-SA: [arabic_SA_chartokenizer]
  ar-SY: [arabic_SY_chartokenizer]

Copilot

Pull request overview

This PR extends NeMo’s TTS tokenizer/G2P support by adding an Arabic character-level tokenizer and adjusting Japanese Katakana pitch-accent G2P behavior for mixed Japanese/English (loanword) text.

Changes:

Added ArabicCharsTokenizer (character-based) and Arabic locale punctuation/grapheme support (ar-MSA).
Updated JapaneseKatakanaAccentG2p handling of ASCII/katakana decisions and unknown symbols/punctuation.
Updated tokenizer unit tests to cover Arabic chars and mixed Japanese/English katakana-accent output.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.

File	Description
`tests/collections/common/tokenizers/text_to_speech/test_tts_tokenizers.py`	Replaces the Japanese katakana-accent test and adds an Arabic chars tokenizer unit test.
`nemo/collections/tts/g2p/models/ja_jp_ipa.py`	Refines ASCII/loanword vs. Latin handling and adds warnings/space replacement for unknown symbols.
`nemo/collections/common/tokenizers/text_to_speech/tts_tokenizers.py`	Introduces `ArabicCharsTokenizer` implementation.
`nemo/collections/common/tokenizers/text_to_speech/ipa_lexicon.py`	Adds `ar-MSA` locale entry, Arabic grapheme set, and locale-specific punctuation.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

XuesongYang · 2026-04-20T17:45:55Z

thanks for adding new tokenizers.

Regarding 4 arabic locales, we decided to use separate tokenizers for ar-SY, ar-AE, ar-MSA, and ar-SA. In this PR, I only found ArabicCharsTokenizer was defined. Could you pls add tokenizers below?
  ar-AE: [arabic_AE_chartokenizer]
  ar-MSA: [arabic_MSA_chartokenizer]
  ar-SA: [arabic_SA_chartokenizer]
  ar-SY: [arabic_SY_chartokenizer]

discussed offline. No need to add one tokenizer per locale.

XuesongYang

lgtm in genral, but need some changes before merging. @quapham pls have a look and take actions accordingly.

i am not familiar either languages. But pls let me know if I am wrong. Thanks!

Unblock from my end

XuesongYang · 2026-04-24T05:32:29Z

@blisc let's pause merging this one now, and merge #15567 first. I believe there are some merging conflicts to address.

Apply isort and black reformatting Fix Hindi chartokenizer, use 'case=upper' and prevent duplicate spaces in the Japanese G2P fallback paths. Apply isort and black reformatting Fix HindiCharsTokenizer backward compat and add Arabic dialect tests Apply isort and black reformatting Add Arabic tokenizer test coverage: diacritics, dialects, punctuation, unknown chars Expand Arabic tokenizer tests: parametrize diacritics, dialects Apply isort and black reformatting added comprehensive test coverage. fix: add back-compatibility, case=mixed, ascii_letters. fix: add charset_version to Hindi/Arabic tokenizers for backward compatibility Introduce a parameter in HindiCharsTokenizer and ArabicCharsTokenizer so old models (v1: case='mixed') keep working while new models train with the corrected charset (v2: case='upper'). - Define CASELESS_SCRIPT_TOKENIZER_TARGETS and DEFAULT_CHARSET_VERSION constants in tts_tokenizers.py - Persist charset_version into the OmegaConf config during training (setup_tokenizers) so .nemo archives record which version was used - Add _migrate_charset_version() helper in magpietts inference utils to pin charset_version=1 for old checkpoints that lack the field, preventing a silent vocabulary mismatch at inference time bugfix: L2_TTS_Fast_dev_runs_Magpietts_OnlineCFGDistillation.sh Signed-off-by: quapham <quapham@users.noreply.github.com> Signed-off-by: Xuesong Yang <1646669+XuesongYang@users.noreply.github.com>

XuesongYang · 2026-04-25T01:08:11Z

@blisc let's pause merging this one now, and merge #15567 first. I believe there are some merging conflicts to address.

all CI passed. @blisc pls review.

blisc · 2026-04-25T15:40:48Z

+            # Persist charset_version so it's saved in .nemo archives and
+            # update_config_for_inference can distinguish old checkpoints
+            # (missing charset_version → v1) from new ones.
+            if (
+                hasattr(tokenizer_config, '_target_')
+                and tokenizer_config._target_ in CASELESS_SCRIPT_TOKENIZER_TARGETS
+                and not hasattr(tokenizer_config, 'charset_version')
+            ):
+                with open_dict(all_tokenizers_config):
+                    tokenizer_config.charset_version = DEFAULT_CHARSET_VERSION


I'm ok to merge this as this. But in the future, we need to refactor this default version to be part of the class not a top level global in the file.

github-actions Bot added TTS common labels Apr 16, 2026

XuesongYang requested review from XuesongYang and Copilot April 20, 2026 17:38

Copilot started reviewing on behalf of XuesongYang April 20, 2026 17:39 View session

Copilot AI reviewed Apr 20, 2026

View reviewed changes

XuesongYang added the Run CICD label Apr 20, 2026

XuesongYang temporarily deployed to test April 20, 2026 17:53 — with GitHub Actions Inactive