search

tencent-ailab / bddm

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

Any way to reproduce your Table 6 #2

Closed CS123n closed 2 years ago

CS123n commented 2 years ago

Hi, This is very interesting work! I find that you report pretty good FID results on CIFAR10. I want to know if this repo is the code of audio generation only or if it can be used to reproduce Table 6, too.

MaxInGaussian commented 2 years ago

Thanks for your interest. This repo only provides the implementation for the audio generation task. I used a separate code base (developed upon ermongroup/ddim) for image generation, though the training and inference procedures for the schedule network in BDDM are pretty much the same as what is provided here.