Open luuksuijkerbuijk opened 6 years ago
I trained Tacotron to 140K steps with the LJSpeech data. That is working great for me.
Now I want to improve the speech by using wavenet. I start a training over the synthesise data created with my Tacotron model.
But when I look at my tensorboard it's not looking good. The output isn't good either.
Anyone with some tips?
Tensorboard:
Eval-dir plot at 130k:
It‘s normal, keep training.
I trained Tacotron to 140K steps with the LJSpeech data. That is working great for me.
Now I want to improve the speech by using wavenet. I start a training over the synthesise data created with my Tacotron model.
But when I look at my tensorboard it's not looking good. The output isn't good either.
Anyone with some tips?
Tensorboard:
Eval-dir plot at 130k: