Vietnamese frontend provenance

The Vietnamese frontend is a native clean-room implementation for the vi-vn Northern/Hanoi-oriented broad phonemic profile. The implementation did not inspect, copy, translate, or depend on hoang1007/vig2p. The gold cases are independently constructed from ordinary Vietnamese vocabulary and the sources below.

Unicode and tone extraction

Sources:

Unicode Standard, Chapter 7: https://unicode.org/versions/Unicode16.0.0/core-spec/chapter-7/
Vietnamese orthography uses U+0300 grave, U+0309 hook above, U+0303 tilde, U+0301 acute, and U+0323 dot below for the five marked tones.

Reason: canonical NFC/NFD normalization and combining-mark classification.

Syllable structure

Sources:

Vietnamese syllable structure summary: https://pmc.ncbi.nlm.nih.gov/articles/PMC9131412/
Luong, Vietnamese Speech Synthesis: https://www.isca-archive.org/iscslp_2006/luong06_iscslp.pdf

Reason: optional onset, medial glide, obligatory nucleus, optional coda, and lexical tone represented as separate structural fields.

Inventory and tones

Source:

ASHA Vietnamese phonology overview: https://pubs.asha.org/doi/10.1044/2023_JSLHR-21-00669
Northern Vietnamese tone description: https://www.frontiersin.org/journals/education/articles/10.3389/feduc.2024.1411660/full

Reason: broad Northern onset, vowel, coda, glide, and six-tone categories. The output deliberately remains broad rather than claiming narrow allophonic accuracy.

Model adaptation

Source:

Kokoro 82M vocabulary is read from this repository’s existing kokorog2p/data/kokoro_config.json.

Reason: Vietnamese rendering is checked against the selected Kokoro model vocabulary. Unsupported characters are reported instead of silently removed.

Gold data

tests/data/vi_gold.json contains independently selected, human-reviewable examples. Expected output is structural broad phonology rendered with the named tone markers. External G2P implementations are not used to generate the curated expected values.