External lexicons and release policy

KokoroG2P consumes lexicons provisioned by Lexphon. It does not ship source dictionaries, pack .g2lex assets, generate registries, or validate producer release files.

English and French

English exposes a default gold lexicon and an explicit lexhint source backed by external assets: the two sources use different encodings but share the KokoroG2P English realization layer.

Language	Lexphon ID	Default	Encoding
US English	`en-us:gold`	yes	`kokoro-v1`
US English	`en-us:lexhint`	no	IPA
GB English	`en-gb:gold`	yes	`kokoro-v1`
GB English	`en-gb:lexhint`	no	IPA
French	`fr-fr:gold`	yes	`kokoro-v1`

Provision released data before using the default English or French dictionary path:

lexphon data install en-us:gold en-us:lexhint en-gb:gold en-gb:lexhint fr-fr:gold
lexphon data verify en-us:gold en-us:lexhint en-gb:gold en-gb:lexhint fr-fr:gold

get_g2p("en-us"), get_g2p("en-gb"), and get_g2p("fr-fr") select gold by default. Use lexicons="lexhint" for the IPA source and lexicons=() for fallback-only operation. silver is not an English runtime option.

Runtime lookup is offline. KokoroG2P does not fetch catalogs, download assets, invoke the Lexphon CLI, or rebuild source dictionaries during construction or lookup. Missing data produces an installation and verification command in the error message.

Released Lexphon catalog entries are the source of truth for these IDs. Install and verify the requested layers before integration or dictionary-backed release checks. KokoroG2P does not restore local copies when a catalog entry or asset is unavailable.

Named German lexicons

German names are application-facing choices backed by externally managed Lexphon IDs:

Name	Lexphon ID	Default
`gold`	`de-de:gold`	yes
`crane`	`de-de:crane`	no
`espeak`	`de-de:espeak`	no
`olaph`	`de-de:olaph`	no
`lexhint`	`de-de:lexhint`	no

Provision them explicitly before using a named dictionary:

lexphon data install de-de:gold
lexphon data verify de-de:gold

Optional layers can be installed when needed:

lexphon data install de-de:crane de-de:espeak de-de:olaph de-de:lexhint
lexphon data verify de-de:crane de-de:espeak de-de:olaph de-de:lexhint

from kokorog2p import available_lexicons, get_g2p

available_lexicons("de")  # ("gold", "crane", "espeak", "olaph", "lexhint")
g2p = get_g2p("de")
g2p = get_g2p("de", lexicons="crane")
g2p = get_g2p("de", lexicons=("gold", "olaph"))

For an explicit selection, the first layer containing a word wins. lexicons="espeak" selects the static de-de:espeak dictionary and is distinct from use_espeak_fallback=True, which is KokoroG2P’s dynamic fallback path.

Distribution policy

KokoroG2P wheels and source distributions contain no migrated .g2lex assets, source lexicons, manifests, generated registries, or lexicon notice bundles. Data provenance and release metadata belong to g2lex-data and Lexphon.

Swedish NST lexicon

The Swedish NST lexicon is produced and published by g2lex-data and installed explicitly through Lexphon. Swedish rules remain the default.

lexphon data install sv-se:nst
lexphon data verify sv-se:nst