Closed buscon closed 5 years ago
Having same problem here. I wish to train a new model with self-collected dataset (in English Language). Can anyone share what is the things we should lookout when training a new model ?
There's some information to get started here: https://github.com/keithito/tacotron/blob/master/TRAINING_DATA.md
Basically collect some speech data (preferably >20 hours), label it, and then train. It's easiest if you use the same format as LJ Speech, but you can also write a preprocessor to handle whatever format you want. For non-English data, see the section in the document linked above.
hi,
it might be a trivial question, but how can I train a new model?