search

cmusphinx / g2p-seq2seq

G2P with Tensorflow
Other
667 stars 196 forks source link

Dealing with words with multiple pronunciations #193

Open ivancarapinha opened 4 years ago

ivancarapinha commented 4 years ago

Hello,

Since this g2p transformer performs phonetic transcription word by word, how does it select the correct pronunciation for a word that has several possible pronunciations? This is very common for many nouns and verbs, for example, the noun "content" and the verb "to content" (to satisfy).

Thank you

nshmyrev commented 4 years ago

It supports n-best output in theory. As for using part of speech as input feature for training, it is also possible, but requires work on model architecture, and, correspondingly, code.

widdiot commented 3 years ago

Does that mean , as of now, for training the g2p model, input dictionary should only have 1-best pronunciations? If not, how to handle multiple pronunciations in the training dictionary?