Dex-TTS checkpoint

[tensimixt]

Hi there, is there a pretrained weights checkpoint for DEX-TTS? Looking forward to trying it out!

Thank you for your great great work!!

[winddori2002]

Hi,

I will upload the checkpoint in near future. (after new recipes for multi-gpu and preprocessing)

Thanks for your interest.

[tensimixt]

Hi,

I will upload the checkpoint in near future. (after new recipes for multi-gpu and preprocessing)

Thanks for your interest.

Hi Thank you for that! Does running through the steps in the below link, lead to exactly the same checkpoint that you are producing?
https://github.com/winddori2002/DEX-TTS/tree/main/DEX-TTS

[tensimixt]

For training GeDEX-TTS is there a max number of speakers that can be in a dataset? is this like 100,1000,10000 speakers?
ideally how many samples do you need to achieve good results per speaker? Assuming each sample is about 6 seconds in length. @winddori2002

Thank you!

[winddori2002]

Hi,

I tested on the VCTK dataset which consists of 100 speakers, and it works fine.

If the number of speakers increases, the model size should be increased.

Considering the VCTK or LibriTTS dataset, more than 30 minutes per speaker is necessary.

Thanks.