search

tencent-ailab / bddm

BDDM: Bilateral Denoising Diffusion Models for Fast and High-Quality Speech Synthesis
Apache License 2.0
218 stars 30 forks source link

VCTK speech recipe #3

Closed jerrymatjila closed 2 years ago

jerrymatjila commented 2 years ago

What is the recipe for reproducing VCTK speech results?

MaxInGaussian commented 2 years ago

The network architecture used in the VCTK experiment were the same. We can similarly train a BDDM after modifying the "train_data_dir", "valid_data_dir" & "sampling_rate" in egs/lj/conf.yml. For the ease of reproduction, I will add egs/vctk in the next commit, including the pre-trained models for VCTK.

jerrymatjila commented 2 years ago

Thanks. Pre-trained models for VCTK will be highly appreciated.

MaxInGaussian commented 2 years ago

You can now find the pretrained models for VCTK at egs/vctk in the latest commit. Thanks.