Logo Lanfrica
  • Home
  • Atlas
  • Insights
  • Docs
  • Sign in

© 2026 Lanfrica. All rights reserved. All copyrights of the resources shown on the Lanfrica website belong to the original copyright holders, unless explicitly stated otherwise.

GhanaNLP/ghana-english-g2p

Domain:

natural language processing

Record type:

software
Creator:
Gha
Host:
Ghanaian English grapheme-to-phoneme: a 104k-word Ghanaian pronunciation lexicon, with espeak-ng behind it for out-of-vocabulary words. # ghana-english-g2p Ghanaian English text to IPA. A **104,623-word Ghanaian pronunciation lexicon** is consulted first; `espeak-english` handles whatever is left over. ```bash pip install ghana-english-g2p ``` ```python from ghana_english_g2p import GhanaEnglishG2P g2p = GhanaEnglishG2P() g2p.ipa("Kwabena went to Achimota") # 'kwabɪna wɛnt tuː atʃimota' g2p.ipa("Kwabena went to Achimota", sep=" ") # 'k w a b ɪ n a w ɛ n t t uː a tʃ i m o t a' ``` ```bash ghana-english-g2p "The Bank of Ghana raised the policy rate." ``` ## Why this exists espeak reads Ghanaian names as if they were English spelling. It does not fail loudly — it returns a confident, fluent, wrong pronunciation: | word | espeak alone | this lexicon | |---|---|---| | Kwabena | `k w e ɪ b n ə` | `k w a b ɪ n a` | | Achimota | `ɐ tʃ ɪ m o ʊ ɾ ə` | `a tʃ i m o t a` | | Okuapenhene | `o ʊ k j uː e ɪ p ə n h iː n` | `o k w a p ɛ n h ɛ n ɛ` | | Bawumia | `b æ w uː m i ə` | `b a w u m i a` | | Twumasi | `t w uː m ɑː s i` | `t w u m a s i` | | Nyantakyi | `n a ɪ ɐ n t æ k ɪ i` | `ɲ a n t a tɕ i` | Every one of these is a name a Ghanaian TTS or ASR system will meet on its first day. The lexicon is what makes the output Ghanaian; espeak is the general-purpose fallback behind it, not the other way round. Rule-based conversion is not an option for this language pair, and that is measurable. Running `ghana-g2p`'s Twi rules over 4,000 words of this lexicon gives **7.6% exact-match and 50.1% phoneme error** — Ghanaian names come out roughly right, English words are destroyed (`concerts` → `tʃ o n tʃ e ɾ t s`). Twi has a shallow orthography and rules suit it; English does not, which is the reason pronunciation dictionaries exist at all. ## One notation, both paths espeak emits stress marks and a run-together transcription; the lexicon stores spaced phones. Both go through the same segmenter, so **you cannot tell from the symbols which path produced a word**: ```python from ghana_english_g2p import segmen …

Visit

github.com

Tasks

text normalization

Languages

AkanBwamu, CwiDinka, SoutheasternTwi

Licenses

MIT