Language support

KokoroG2P provides language-specific phonological frontends over a common prepared-text pipeline. Select the language explicitly with phonemize_prepared() or get_g2p(). Written-to-spoken semantic preparation is outside every frontend.

Common contract

from kokorog2p import phonemize_prepared

result = phonemize_prepared("Bonjour le monde", language="fr")

All frontends preserve prepared input and expose token-level phonemes. Numbers, quantities, dates, times, currencies, units, abbreviations, and URLs must be prepared by the calling application. Language switching is also explicit: use OverrideSpan or annotation language metadata for a known foreign span. No generic automatic detector is included.

Native and dictionary frontends

Language	Code	Frontend
English (US/GB)	`en-us`, `en-gb`	dictionary and rule-based
German	`de`	dictionary and rule-based
French	`fr`	dictionary and rule-based
Spanish	`es`	rule-based
Italian	`it`	rule-based and lexicon
Portuguese	`pt-br`, `pt-pt`	rule-based
Czech	`cs`	rule-based
Vietnamese	`vi`, `vi-vn`	native rule-based
Swedish	`sv-se`	native rule-based
Russian	`ru`	LexHint with eSpeak fallback
Kazakh	`kk`	eSpeak adapter
Hindi	`hi`, `hi-in`	eSpeak-NG raw-IPA adapter
Hebrew	`he`	Phonikud adapter
Arabic	`ar`	optional diacritizer adapter
Chinese	`zh`	pypinyin/Zhuyin frontend
Japanese	`ja`	pyopenjtalk or Cutlet
Korean	`ko`	LexHint fast path plus g2pK
Thai	`th`	LexHint with native fallback

Hindi accepts hi, hi-in, hi_IN, hin, and hindi, and uses the eSpeak-NG hi voice as a lazy raw-IPA adapter for stock Kokoro v1.0. The frontend preserves Hindi vowel length, nasalization, aspiration, retroflexion, and native vowel quality, then validates the result against the model vocabulary. Hindi is prepared-text only, has no bundled or downloaded runtime lexicon, and provides no automatic-routing evidence.

Install language-specific optional dependencies from the matching extras in pyproject.toml, for example pip install "kokorog2p[ja]".

Backend and model controls

Use get_g2p() for backend-specific options such as backend, voice, or an explicitly selected spaCy model. Semantic preparation options and abbreviation registries are not accepted by the v0.9 API.

from kokorog2p import get_g2p

g2p = get_g2p("en-us", use_spacy=False)
print(g2p.phonemize("Hello world"))

Prepared-language composition

A cross-package pipeline may prepare text and then route the prepared result explicitly:

from spokenform import prepare_for_kokorog2p
from kokorog2p import phonemize_prepared

prepared = prepare_for_kokorog2p("Meeting at 2 kg", language="en").spoken_text
result = phonemize_prepared(prepared, language="en-us")

Spokenform is optional and is not imported by the core package.