Fine tune adding of new speaker to existing models tts wavernn

[JRMeyer]

>>> Salisbury_Espinosa
[February 13, 2020, 7:57am]

Is it possible for existing pre-trained models tts and wavernn to add a
new dataset and a new speaker. and retrain the model for this speaker?
if so, how?

[This is an archived TTS discussion thread from discourse.mozilla.org/t/fine-tune-adding-of-new-speaker-to-existing-models-tts-wavernn]

[JRMeyer]

>>> erogol
[February 14, 2020, 2:35pm]

Possible. You can use the universal vocoder model released here
https://github.com/erogol/WaveRNN

To finetune the TTS model for your speaker this might help: slash

{.site-icon width='32'

### mozilla/TTS

forum: https://discourse.mozilla.org/c/tts) - mozilla/TTS

Basically, you need to format your dataset as LJSpeech and resume the
training of one of the released models. Good luck.

[Archived Post]

[JRMeyer]

>>> Vinay_Sharma
[November 2, 2020, 1:59pm]

, is there some colab notebook
to show how to resume the training of one of the released model for new
dataset.


[Archived Post]