Hi, I am a novice in TTS (and machine learning in general) and I apologize in advance if this question is dumb.
I am trying to fine tune the released Tacotron2 DDC model from the wiki. However, I am unable to initialize the downloaded model fully. As shown below, only 85/105 layers were restored.
The above output was obtained with the following settings:
All codes used were from the current master branch e9e0784
The config file was nearly identical to this file , and I only updated the necessary path information.
After some exploring on the past discussions and issues, I tried 2 ways to resolve this problem.
Method 1: Use the right commit version for the released Model
72a6ac5
Result: Similar to the output above.
Method 2: Use the config.json inside the google drive for tacotron2DDC.
Result:
Got this error:
AssertionError: [!] decoder_loss_alpha not defined in config.json
I am quite lost on why I cannot load the complete model, and I hope someone can point me in a direction for further investigation.
Thank you for reading and stay safe.