vig2p

Vietnamese G2P package with an internal pure-Python port of the sea-g2p 0.7.6 backend and the Kokoro-compatible adapter used by the Vietnamese Kokoro fine-tuning workflow.

Install

pip install vig2p

For local development:

python -m pip install -e ".[dev]"

CLI

vig2p "tường nhà khách. thường nhà khách."
python -m vig2p "trước chước số xố giải dải"

Read one utterance per line:

vig2p --file metadata_texts.txt

JSON output:

vig2p --json "tường nhà khách"

Python

from vig2p import VietnameseG2P, phonemize_text

print(phonemize_text("tường nhà khách"))

g2p = VietnameseG2P()
print(g2p("trước chước số xố giải dải"))

Adapter Policy

• tʃ -> ʧ
• t̪ -> t
• source words starting th rewrite first raw t -> θ, including unaccented Vietnamese words like theo
• source words starting tr rewrite first raw ʧ -> ʈʂ
• source words starting s rewrite first raw s -> ʂ, including unaccented Vietnamese words like sinh; obvious non-Vietnamese s clusters such as st/sp/sk/... are left alone
• source words starting gi/gì variants rewrite first raw z -> ʝ
• e- -> æ
• 1/7 -> →, 2 -> ↘, 3/ɜ -> ↗, 4 -> ↓, 5 -> ʔ↗, 6 -> ʔ↓
• ɗ -> d, ʐ -> ʒ, đ -> d
• curly apostrophes behave like apostrophes
• – -> —
• unsupported *, /, & are stripped or spaced before training phonemes

Backend

vig2p vendors the Apache-2.0 sea-g2p 0.7.6 dictionary as package data and uses a pure-Python reader/phonemizer in vig2p._sea_g2p. It does not depend on the external sea-g2p Python package or its Rust extension at runtime.

Test

python -m unittest discover -s tests -v

Publish

Build and check the package:

python -m build
python -m twine check dist/*

Upload to TestPyPI first:

python -m twine upload --repository testpypi dist/*

Then upload to PyPI:

python -m twine upload dist/*