https://uberduck.ai/ logo
Powered by
# machine-learning
  • j
    Output result
  • t
    Neat
  • g
    He's trained to 250 epochs on 300 clips worth of data.
  • g
    He sounds a bit undertrained though.
  • u
    Are you trying to make the voice more accurate? Like, making it sound so real?
  • g
    Yeah. I just resumed its training in Colab, and it seems it only got to 186, or only saved at 186.
  • g
    Kaggle was bugging out last night for me.
  • g
    He's definitely stonking here.
  • m
    Best stonks I’ve seen for now
  • s
    beautiful stonkin'
  • g
    He still has a way to go though. There was some high validation loss on him.
  • g
    That spiked when I took a shower, but it's going back down now.
  • e
    That sounds pretty good to me!
  • g
    Yeah, I know Prime is kinda monotone, but this isn't the right kind of monotone.
  • t
    @User Taking a look at the notebook right now Running into lovely issues
  • h
    sounds fun
  • t
    No such file or directory: '/content/datasetunzip/__MACOSX/label.json'
  • h
    > __macosx
  • t
    When I select /mlp/ dataset
  • t
    Yeah, I
  • h
    oh because it's
  • h
    it's made for single-character datasets
  • t
    I can't see anything in the code for it
  • h
    specifically the .tar files from sortanon's drive
  • h
    or no wait hm
  • h
    right, it assumes the hidden __macosx folder is a character
  • h
    -nono wait
  • h
    a voice clip
  • h
    yes
  • t
    Ought I change the directory listed in the code?
1...655656657...1068Latest