PhonoGlyphe is a G2P (grapheme to phoneme) 6M transformer model meant as a fallback method for the Misaki G2P engine. It currently only supports the English language.
from misaki import en
from PhonoGlyphe import G2PModel
fallback = G2PModel(device="cpu") # Note: with its small size, PhonoGlyphe is often faster on CPU
g2p = en.G2P(trf=False, british=False, fallback=fallback)
text = "Misaki is a G2P engine designed for Kokoro models."
phonemes, tokens = g2p(text)
print(f"Phonemes: {phonemes}")You can easily listen to the final result using [.](/<GENERATED PHONEMES>/) in this HF Space.
Phonoglyphe was trained using the English dictionaries of the Misaki project. The training code can be found in the training folder.
Thanks to Noelle for her support and all her precious advice.