search

rhasspy / larynx

End to end text to speech system using gruut and onnx
MIT License
824 stars 49 forks source link

German voice #4

Closed alt131 closed 3 years ago

alt131 commented 3 years ago

Did you think about using pavoque-data for German voice? https://github.com/marytts/pavoque-data/releases/tag/v0.2 It has better quality in comparing with thorsten audio dataset. It has restriction for commercial using but as I see you already have a voice with that restriction.

synesthesiam commented 3 years ago

I did train a voice with it, actually :slightly_smiling_face:

It didn't sound very good, and I now know that I need to do a forced alignment. I just finished training a German Kaldi model this past weekend with the same IPA phonemes, so I will be aligning and training this voice this week :+1:

alt131 commented 3 years ago

If you already tried it you can close this issue because I didn't have the hardware to train the model by myself and check it in practice.

synesthesiam commented 3 years ago

I'll keep the issue open for now, since I'll be training it soon. Do you know of any good TTS datasets? Or are you interested in contributing your voice?

alt131 commented 3 years ago

I'm afraid I say no in both cases but if I find TTS datasets with good quality I will let you know.

synesthesiam commented 3 years ago

I successfully trained this and three other German voices :slightly_smiling_face:

https://rhasspy.github.io/larynx/#de-de