tugaphone — dialect-aware Portuguese phonemizer

tugaphone converts Portuguese text to IPA across all five Lusophone dialect groups. It combines a curated phonetic lexicon, part-of-speech tagging for homograph disambiguation, meaning-based heterophone resolution via bifonia, and a scientifically-grounded regional-accent layer.

O gato dorme.
pt-PT → ˈu gˈa·tu ˈdoɾ·mɨ ˈ···
pt-BR → ˈu gˈa·tʊ ˈdoɾ·mɪ ˈ···
pt-AO → ˈu gˈa·tʊ ˈdoɾ·me ˈ···
pt-MZ → ˈu gˈa·tu ˈdoɾ·me ˈ···
pt-TL → ˈu gˈa·tʊ ˈdoɾ·me ˈ···

Install

pip install tugaphone

30-second quick start

from tugaphone import TugaPhonemizer

ph = TugaPhonemizer()
print(ph.phonemize_sentence("O gato dorme.", "pt-PT"))
# ˈu gˈa·tu ˈdoɾ·mɨ ˈ···

TugaPhonemizer() loads the lexicon and POS tagger once; then call phonemize_sentence(text, lang) as many times as you like. Output is a space-separated phoneme string — one token per word — with ˈ marking primary stress and · marking syllable boundaries.

Features

Five dialect inventories

for code in ["pt-PT", "pt-BR", "pt-AO", "pt-MZ", "pt-TL"]:
    print(code, "→", ph.phonemize_sentence("Choveu muito ontem.", code))
# pt-PT → ʃu·ˈvew mˈũj·tu ˈõ·tẽ ˈ···
# pt-BR → ʃo·ˈvew mwˈĩ·tʊ ˈõ·tẽ ˈ···
# pt-AO → ʃo·ˈvew mˈũjn·tʊ ˈõ·tẽ ˈ···
# pt-MZ → ʃu·ˈvew mˈũj·tu ˈõ·tẽ ˈ···
# pt-TL → ʃo·ˈvew mˈuj·tʊ ˈõ·tẽ ˈ···

Homograph disambiguation

Heterophonic homographs are resolved at two levels:

1. Meaning-based (via bifonia): sede thirst vs HQ, forma mould vs shape.
2. POS-based: gosto noun /ˈgoʃtu/ vs verb /ˈgɔʃtu/, para preposition vs verb.

print(ph.phonemize_sentence("Eu gosto de música."))   # verb → ˈgɔʃ·tu
print(ph.phonemize_sentence("Tenho bom gosto."))      # noun → ˈgoʃ·tu

Sub-regional accents

RegionalTransforms presets layer phonological rules on top of any dialect. Rules are grounded in published phonology (Cintra 1971; ALEPG). Every preset is reachable by its BCP-47 private-use code:

# Porto: stressed /o/ → [uo] (rising diphthong)
print(ph.phonemize_sentence("O vinho é muito bom.", "pt-PT-x-porto"))
# ˈu bˈi·ɲu ˈɛ mˈũj·tu bˈuõ ˈ···

# Açores: stressed /u/ → [y], l-palatalisation
print(ph.phonemize_sentence("O vinho é muito bom.", "pt-PT-x-azores"))
# ˈy vˈi·ɲu ˈɛ mˈỹj·tu bˈõ ˈ···

from tugaphone import list_dialects
print(list_dialects())   # all 20 registered codes

Available presets: NorthernDialect, PortoDialect, MinhoDialect, BragaDialect, FamalicaoDialect, FafeDialect, TrasMontanoDialect, CoimbraDialect, AlentejoDialect, AlgarveDialect, MadeiraDialect, AzoresDialect — importable from tugaphone.regional and passable as regional_dialect=, which overrides the code-derived preset.

Number normalization

Digits are spelled out with gender agreement and long/short scale per dialect:

from tugaphone.number_utils import normalize_numbers

normalize_numbers("vou comprar 1 casa")   # 'vou comprar uma casa'
normalize_numbers("vou adotar 1 cão")    # 'vou adotar um cão'
normalize_numbers("comprei 2 casas")     # 'comprei duas casas'

Syllabification and stress

Syllabification is handled by silabificador, registered as an orthography2ipa syllabifier plugin. Stress detection delegates to orthography2ipa's declarative StressRules.

Rules-only mode

Pass an IRREGULAR_WORDS-emptied dialect inventory to bypass the lexicon and use only grapheme rules — useful for testing rule coverage or synthesising unknown words.

orthography2ipa plugin interface

TugaphoneG2PPlugin implements orthography2ipa's G2PPlugin interface; SilabificadorSyllabifier implements its SyllabifierPlugin interface and is registered at the orthography2ipa.syllabify entry point.

from tugaphone.plugin import TugaphoneG2PPlugin

p = TugaphoneG2PPlugin(lang="pt-BR")
print(p.transcribe("o gato dorme"))   # ˈu gˈa·tʊ ˈdoɾ·mɪ

Sibling libraries

tugaphone is part of the TigreGotico Portuguese NLP stack:

Documentation

• docs/quickstart.md — install, first call, dialect overview
• docs/dialects.md — five inventories and sub-regional accent presets
• docs/homographs.md — meaning-based and POS-based disambiguation
• docs/numbers.md — number normalization and gender agreement
• docs/api.md — full class and function reference
• docs/tokenizer.md — the Sentence → Word → Grapheme → Character model
• docs/advanced.md — accents, serialization, integration

License

Apache License 2.0. See LICENSE.