Begin with a new installation of MFA, and head to the directory. First run validate with the original dictionary.
mfa validate ~/Downloads/tb/my_corpus english_us_arpa english_us_arpa
We see that there is in deed an section of corpus that is out-of-vocab.
INFO - 11 OOV word types
INFO - 18 total OOV tokens
Therefore, we will generate a new dictionary based on the existing dictionary of english_us_arpa.
First download the english_us_arpa model
mfa model download g2p english_us_arpa
Then, perform the actual dictionary generation:
mfa g2p english_us_arpa ~/Downloads/tb/my_corpus ~/Downloads/tb/my_corpus/new_dict.txt
There is a chance this command fails with
There was an issue importing Pynini, please ensure that it is installed. If you are on Windows, please use the Windows Subsystem for Linux to use g2p functionality.
If so, install pynini
conda add pynini
Finally, run the mfa g2p command above to generate pronunciations.
You should end up with a file named new_dict.txt, which should include missing words.
Finally, perform alignment with this new dictionary.
mfa align ~/Downloads/tb/my_corpus ~/Downloads/tb/my_corpus/new_dict.txt english_us_arpa ~/Downloads/tb/my_corpus_output
Notice here the second argument of mfa align is no longer english_us_arpa, our base dictionary. Instead, it is our custom dictionary.