Kyubyong Park

Actually nothing but one: replacing single phonemes with triphones. I know it's not enough for this task, so I'm looking for a better architecture/model.

No for training always English speech samples are uswd. For inference different languages are to be tested.

Exactly.

@tetelestia Thanks! I corrected @puentev Would you check again? To me there seems no problem.

Oh I'm very sorry. Would you check again?

Hi, I'm preparing single speaker datasets and pretrained models for 10 languages with Tommy and we're going to release them soon. I'll keep you posted.

The best suggestion I can do is to consider other models like tacotron or dctts. You can find their implementation in my github. They are much easier to train.

Sure. Python 2 TF 1.3 linux

Hi Shuming, Amazing! Actually I have had a quite similar idea, so your work is particularly impressive! Kyubyong 2018-01-19 16:48 GMT+09:00 Shuming Fang : > Hi, Kyubyong. Based on your...

Did you try with Python 2?