Skip to content

Conversation

@phuetz
Copy link

@phuetz phuetz commented Dec 28, 2025

Add comprehensive French language support to Misaki G2P:

Features

  • French G2P module (misaki/fr.py)
  • Pronunciation dictionary with 15,000+ entries from Lexique.org (fr_gold.json)
  • Phoneme documentation (FR_PHONES.md)
  • Abbreviation expansion (M., Mme, Dr, etc.)
  • Number-to-words conversion (up to billions)
  • Time expression handling (14h30 → "quatorze heures trente")
  • Ordinal expansion (1er → "premier")
  • espeak-ng fallback for unknown words

Fixes common espeak-ng pronunciation issues

  • Imparfait endings (-ait/-ais): était → /etɛ/
  • Words with "ai" sound: commissaire → /kɔmisɛʁ/
  • Nasal vowels: main → /mɛ̃/

Data sources

  • Lexique.org - 15,000 most frequent French words
  • Manual corrections for mispronounced words

Author: Patrice HUETZ

@phuetz phuetz force-pushed the feat/french-support branch 2 times, most recently from eb1cb48 to 635b20d Compare December 28, 2025 19:57
Add comprehensive French language support to Misaki G2P:

Features:
- French G2P module (misaki/fr.py)
- Pronunciation dictionary with 15,000+ entries from Lexique.org (fr_gold.json)
- Phoneme documentation (FR_PHONES.md)
- Abbreviation expansion (M., Mme, Dr, etc.)
- Number-to-words conversion (up to billions)
- Time expression handling (14h30 → "quatorze heures trente")
- Ordinal expansion (1er → "premier")
- espeak-ng fallback for unknown words

Fixes common espeak-ng pronunciation issues:
- Imparfait endings (-ait/-ais): était → /etɛ/
- Words with "ai" sound: commissaire → /kɔmisɛʁ/
- Nasal vowels: main → /mɛ̃/

Data sources:
- Lexique.org (15,000 most frequent French words)
- Manual corrections for mispronounced words

Author: Patrice HUETZ
@phuetz phuetz force-pushed the feat/french-support branch from 635b20d to 32d490a Compare December 28, 2025 20:00
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant