sp-uhh / sgmse

Score-based Generative Models (Diffusion Models) for Speech Enhancement and Dereverberation
MIT License
521 stars 76 forks source link

Network conditioning #25

Open philgzl opened 1 year ago

philgzl commented 1 year ago

Song conditions the network on the standard deviation $\sigma(t)$ of the perturbed variable as per here. This is in line with the framework layed by Karras. However it seems to me that the network here is conditioned on the time $t$ instead (here). Is this intentional?