Jèrriais phonemizer

A phonemizer for Jèrriais (Jersey Norman), written in Snowball. The rules in src/*.sbl are the source; Snowball compiles them to Python, JavaScript and C. The output suits TTS and alignment: affricates are single symbols (ʧ, ʤ), and long vowels and diphthongs are marked with ː (eːi, oːu).

Python

pip install jerriais-phonemizer

from jerriais_phonemizer import phonemize

phonemize("bouônjour, bouonne gens.")       # 'bwõːʒur bwɔn ʒɑ̃'
phonemize("bouônjour, bouonne gens.", "|")  # 'b|w|õː|ʒ|u|r b|w|ɔ|n ʒ|ɑ̃'

Input should be lowercase: uppercase letters are dropped, like punctuation. Nasal vowels come out as a letter plus U+0303 (combining tilde), never precomposed (õ).

JavaScript and C

make javascript writes dist/javascript/phonemizer.js (the class is the default export; call new Phonemizer().phonemize(text)), and make c_example builds a command-line tool, dist/c/phonemizewords, that phonemizes a file of words or phrases one per line. These aren't published as packages.

Building

make fetches and builds Snowball at the release pinned in the Makefile (SNOWBALL_VERSION), then compiles the rules for every language. The Python package's generated files (python/jerriais_phonemizer/_snowball/) are committed, so after changing a .sbl file run make python and commit the result.

• make test: the Python tests (tests/test_cases.csv, plus the two snapshots).
• make snapshot: regenerate tests/snapshot.dic (the spellchecker's word list) and tests/phrases.dic from the C build. Do this only to accept a rule change, and read the diff.

Releasing

Bump version in pyproject.toml, commit, tag (vX.Y.Z) and push the tag. Then run the "Publish to PyPI" workflow (Actions tab) with that tag: it builds the wheel and sdist from the tag and uploads them to PyPI by trusted publishing.

Test data

• tests/jerriais.dic: the Jèrriais spellchecker word list from the Société Jersiaise's spellchecker page, one word per line. tests/snapshot.dic is its phonemization.
• tests/phrases.txt: example phrases. tests/phrases.dic is their phonemization.

The sdist leaves out the word list and its snapshot (3.6 MB); the test that uses them skips without them.

Licence

GNU Affero General Public License, version 3 or (at your option) any later version (AGPL-3.0-or-later, LICENSE). If you run a modified version as a network service, the AGPL asks you to offer its source to the service's users. The Snowball runtime files in python/jerriais_phonemizer/_snowball/ are under Snowball's BSD licence (COPYING there).

For alternative licence terms, contact@jerriais.tech