tencent-ailab / bddm

BDDM: Bilateral Denoising Diffusion Models for Fast and High-Quality Speech Synthesis

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

.mel Data and Scoring Network

LaughingC0ffin opened this issue · comments

Hello fellas,

im currently working on a project and try to make your model working.

My first question or issue is your usage of .mel files. Im not having any clue how they get generated. Or from what. My first guess would be sth like Tacotron 2 or sth like that. Since now i was always working with .npy files and im having a really hard time with .mel. Could you tell me where they come from and how they get generated? Would be enough for me to know with with extension they get saved to reproduce that for my data.

Second one is the Score Network. Do you have any references how you train this and where this comes from?

Best regards

LaughingCoffing

Okay first one is deprecated. I figured it out. You just use torch.save and give it a fancy extension name. Thats insane.

RuntimeError: NCCL error in: /pytorch/torch/lib/c10d/ProcessGroupNCCL.cpp:825, unhandled system error, NCCL version 2.7.8

Its getting worse here.

After spamming different Pytorch and Cuda Versions is working.