Can not synthesis a good voice file

[Hayes515]

Using LJSpeech-1.1 datasets.
Avoid getting OOM error, I just changed less in the hparams.py file("wavenet_batch_size = 2 " and "train_with_GTA = False" )

After training Tacotron with 100k steps ,I got these below .

And I trained wavenet with 80k steps , got step-80000-waveplot in /logs-Tacotron-2/plots.

The wavs in /logs-Tacotron-2/eval-dir/wavs are noisy.
step-80000-wav.tar.gz

I tried to run python synthesize.py to synthesize new sounds, they were noisy too.
Do I need to train wavenet more steps ? Is these something I miss doing?

[Yeongtae]

Check my report #183 #156

[Thien223]

from the plot images, I think there is no problem with your wavenet. Try training more time steps.

[zhjy8827]

@Hayes515

Hello, I have the same problem as you... Can you solve it?