https://uberduck.ai/ logo
Powered by
# machine-learning
  • c
    I'm 100% sure it is applied
  • c
    Probably mega b can explain this better though I don't want to ping him
  • h
    maybe the supersampling step was emphasizing some flaws? try setting the 4 in the source code of the last cell to 0
  • c
    Tried that too, I'm using the gpu version
  • c
    Lemme see if I have an example here
  • z
    @Cris140 are you sure that the generator is in eval mode?
  • c
    duration rising monkaS
  • c
    haha oops i was training on the wrong dataset the whole time
  • u
  • c
    This part?
  • z
    Yeah
  • c
    if this still doesnt work i have a different idea
  • c
    Do I need to assign a value inside the "()"?
  • c
    are you porting hifi to cpu question mark
  • c
    I'm using in cpu mode, yes
  • z
    nah the eval() should do it. that's not the issue
  • c
    oh wait What is. (or was) the issue
  • c
    the robotic sound
  • c
    are you Using: The Pretrained Generator Model
  • c
    Nope
  • c
    It only happens in a few models
  • c
    It works fine in Uberduck but in colab it sounds horrible
  • c
    Could it be some denoiser?
  • u
    its the hifigan
  • c
    he finetuned it though
  • c
    have you tested in eval mode like Zhong Won Fei said
  • u
    didnt someone mention supersampling
  • c
    thats talknet
  • u
    really?
  • u
    swore it was mentioned in regard to taco colab
1...675676677...1068Latest