🇳🇬 Yoruba-G2P

Tone-Aware Yoruba Grapheme-to-Phoneme Toolkit (IPA + ASCII + MFA-Ready)

Yoruba-G2P is a fully deterministic Python package for converting Yorùbá text → phoneme sequences, with correct tones, nasal handling, affricates, and labial-velars.

It outputs:

• ✔ IPA dictionary
• ✔ ASCII-safe dictionary (for MFA, ESPnet, Kaldi)
• ✔ Phoneset file
• ✔ Lexicon statistics
• ✔ CLI + Python API
• ✔ Works on any Yoruba transcript
• ✔ No ML training required → fully rule-based + Epitran-backed

Installation

From PyPI (recommended)

pip install yoruba-g2p

From GitHub

pip install git+https://github.com/<your-username>/yoruba-g2p.git

Quick Start (Python API)

from yoruba_g2p import YorG2P

g2p = YorG2P()
print(g2p.to_ipa("ọ̀yọ́"))
print(g2p.to_ascii("ọ̀mọ́"))

Output:

['ɔ_L', 'j', 'ɔ_H']
['O_L', 'm', 'O_H']

Another example:

g2p.to_ipa("àwọn")

['a_L', 'w', 'ɔ_M', 'n']

Command-line Interface (CLI)

Convert a sentence:

yoruba-g2p --ipa "àwọn ọmọ ń lọ"
yoruba-g2p --ascii "àwọn ọmọ ń lọ"

Build lexicons from .lab transcripts:

yoruba-g2p build-lexicon \
  --lab-dir data/lab/train \
  --out-dir dict/

You will get:

dict/ipa.dict
dict/ascii.dict
dict/phoneset.txt
dict/stats.json

All MFA-ready.

What Yoruba-G2P Produces

1. IPA Dictionary

ọ̀mọ́    ɔ_L m ɔ_H
jẹ́      d͡ʒ ɛ_H
àwọn     a_L w ɔ_M n

2. ASCII Dictionary

ọ̀mọ́    O_L m O_H
jẹ́      dZ e_H
àwọn     a_L w O_M n

3. Phoneset (phoneset.txt)

a_M
a_H
a_L
e_M
e_H
e_L
ɛ_M
ɛ_H
ɛ_L
i_M
i_H
i_L
o_M
o_H
o_L
ɔ_M
ɔ_H
ɔ_L
kp
gb
s
t
d
m
n
n_H
n_L
...

4. Stats (stats.json)

{
  "num_words": 5478,
  "problem_words": 2,
  "phoneset_size": 41
}

Theory Behind Yoruba-G2P

Yoruba tone is lexical, and vowels carry tone:

The engine performs:

1. Unicode-normalization (NFC)

2. Orthographic vowel + tone extraction

3. IPA transliteration via Epitran

4. IPA segmentation (phones: vowels, consonants, digraphs)

5. Tone reattachment using your vowel map

   • Guarantees correct H/M/L tones

6. Handling:

   • nasal vowels (ɛ̃ → ɛ_M + n)
   • syllabic nasals (ń → n_H, ǹ → n_L, decomposed forms)
   • affricates (d͡ʒ, t͡ʃ)
   • labial-velars (kp, gb)

Outputs IPA or ASCII-friendly phones (e.g., kp, gb, dZ).

Directory Outputs (for MFA)

If you run build-lexicon, your output folder will contain:

ipa.dict       # MFA-ready IPA lexicon
ascii.dict     # ASCII-safe lexicon
phoneset.txt   # sorted unique phones
stats.json     # statistics 

Then feed into MFA:

mfa train \
  --corpus wavs/ \
  --dictionary dict/ipa.dict \
  --output align_out/

Demo Notebook

A worked example is provided:

demo_yoruba_g2p.ipynb

It includes:

• IPA & ASCII conversion
• Batch lexicon building
• Phoneset extraction
• Tone distribution visualization
• Handling problem words

Project Structure

yoruba_g2p/
│
├── core.py     # Main G2P engine
├── utils.py    # helper rules, ASCII mapping
├── cli.py      # command line interface
└── __init__.py

Contributing

Pull requests are welcome!

Steps:

1. Fork the repo
2. Create a new branch: git checkout -b feature-name
3. Commit changes
4. Push and open a PR

Issues, feature requests, and Yoruba orthography corrections are welcome.

📣 Citation (for research)

Osakuade, O. (2025). Yoruba-G2P: A tone-aware grapheme-to-phoneme converter for Yorùbá. 
https://github.com/OpeyemiOsakuade/yoruba-g2p

License

MIT License — free for academic and commercial use.

⭐ Support Yoruba NLP

If this toolkit helps you, please give the repo a star ⭐ to promote more open-source resources for African languages.