Closed zhangjiulong closed 7 years ago
Do you want the sequence of phones that the acoustic model would compute without a language model? The same that is computed during validation in the training step?
tokens.txt contains the tokens of the FST, with blank and disambiguation tokens, while units.txt contains the list of phones.
as in the issues98 why units.txt do not contains blank id?
the blank is implicit - you don't specify it, it appears between all phones, but it is not a unit that gets used in the computation of the objective function.
Thanks for your replay, I will try to understand.
whichi file to use when decoding to get phone list.