Dev.ap/phone based models - #27
Conversation
| ) | ||
| from everyvoice.config.type_definitions import DatasetTextRepresentation | ||
| from everyvoice.model.feature_prediction.FastSpeech2_lightning.fs2.prediction_writing_callback import ( | ||
| resolve_chunked_basename, |
There was a problem hiding this comment.
urgh, cross import between submodules!? maybe we should considering refactoring this into everyvoice.base_cli?
|
|
||
|
|
||
| def encode_text_for_inference( | ||
| module, |
There was a problem hiding this comment.
It's a bit confusing to me that the model parameter to this function as well as to synthesize_audio_styletts2 (and other synthesis related functions in styletts2) is called module rather than model. I did some digging and I see they're defined in class StyleTTS2Module itself built on torch and/or lightning modules, as are other model classes. Is there a meaningful semantic difference here or did it just come from the upstream StyleTTS2 repo?
joanise
left a comment
joanise
left a comment
There was a problem hiding this comment.
Generally good. Moving the function that is now shared between fs2 and styletts2 into EV would be nice, but up to you, not required.
The module/model question, up to you if you want to change anything.
1736885 to
5732dc8
Compare
implement same inference time text processing as fs2
same implementation as fs2
5732dc8 to
aacdc79
Compare
2 participants
EveryVoiceTTS/EveryVoice#868