Marxav/frpron

Fr-word to phonemic pronunciation

This model aims at predicting the syllabized phonemic pronunciation of the French words.

The generated pronunciation is:

• A text string made of International Phonetic Alphabet (IPA) characters;
• Phonemic (i.e. remains at the phoneme-level, not deeper);
• Syllabized (i.e. characters '.' and '‿' are used to identify syllabes).

Such pronunciation is used in the French Wiktionary in the {{pron|...|fr}} tag.

To use this model, simply give an input containing the word that you want to translate followed by ":", for example: "bonjour:". It will generate its predicted pronunciation, for example "bɔ̃.ʒuʁ".

This model remains experimental. Additional finetuning is needed for:

• Homographs with different pronunciations,
• French liaisons,
• Roman numerals.

The input length is currently limited to a maximum of 60 letters.

This work is derived from the OTEANN paper and code, which used minGTP.

More information on the model, dataset, hardware, environmental consideration:

The training data

The dataset used for training this models comes from data of the French Wiktionary.

The model

The model is build on gpt2