Training a new G2P model¶
Another tool included with MFA allows you to train a G2P (Grapheme to Phoneme) model automatically from a given pronunciation dictionary. This type of model can be used for generating dictionaries. It requires a pronunciation dictionary with each line consisting of the orthographic transcription followed by the phonetic transcription. The model is generated using the Phonetisaurus software, which generates FST (finite state transducer) files. The G2P model output will be a .zip file like the acoustic model generated from alignment.
To train a model from a pronunciation dictionary, the following command is used:
bin/mfa_train_g2p /path/to/dictionary/file /path/to/output/dictionary
This should be used if there are instances of a single orthographic character corresponding to more than 2 phones (common in Korean hangul, Chinese character orthography, etc.).
Run a validation on the dictionary with 90% of the data as training and 10% as test. It will output the percentage accuracy of pronunciations generated.
See Example 3: Train Mandarin G2P model for an example of how to train a G2P model with a premade toy example.