Closed nmfisher closed 4 years ago
The code of Tacotron-2 with Tensorflow has been synchronized here. Both of them support multi-speaker model. And the preprocessing is consistent with each other. You can extract the ground truth mel files on disk through preprocess.py and then train them both on Torch and Tensorflow version.
By the way the evaluation of multi-speaker model has been released in https://github.com/begeekmyfriend/tacotron2/issues/1
Yes, I understand that the code is compatible. My question was whether you have/will release the trained model weights that you used to create the evaluation samples?
Sorry that the multi-speaker corpus is not public domain, it belongs to business. So I cannot release the pre-trained model. Sorry again about it.
OK, thanks!
I beg your pardon that you might expand worker number of data loader for training acceleration https://github.com/begeekmyfriend/tacotron2/commit/35e7f455372707b3e7146ddda7bdb45f7de5daa5
First off - great work!
Second - will you make any pretrained models available? There are a few scattered throughout various issues in Rayhane's repository, but they all seem to be out of date and incompatible with the current master.