search

TensorSpeech / TensorFlowTTS

:stuck_out_tongue_closed_eyes: TensorFlowTTS: Real-Time State-of-the-art Speech Synthesis for Tensorflow 2 (supported including English, French, Korean, Chinese, German and Easy to adapt for other languages)
https://tensorspeech.github.io/TensorFlowTTS/
Apache License 2.0
3.8k stars 810 forks source link

Support for German Umlaute ÖÄÜöäü and ß. Ignore unnecessary warnings during preprocessing (>= -> >) #709

Closed fkemeth closed 2 years ago

fkemeth commented 2 years ago

In the current implementation, German letters ÖÄÜöäü and ß are not supported. I changed this in this commit, and added support for Tacotron2 training with the correct German alphabet.

Note that without these letters the text-to-speech model cannot pronounce many German words correctly. This will, however, also affect backwards compatibility for models trained on the "old" wrong German alphabet.

I also adjusted a warning condition in the preprocessing script, which raised unjustified warnings due to a greater-equal instead of a just greater condition.

fkemeth commented 2 years ago

Pytest fails due to multiple

tensorflow.python.framework.errors_impl.AlreadyExistsError: Another metric with the same name already exists.

This usually occurs when there are diverging keras and tensorflow versions (see (https://stackoverflow.com/questions/68970841/another-metric-with-the-same-name-already-exists))

This is possible due to the Keras version released in September, but should definitely not be due to this MR.

Any ideas on how to deal with this? Adjusting the requirements?

dathudeptrai commented 2 years ago

@fkemeth Let me check.

stale[bot] commented 2 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.