https://uberduck.ai/ logo
Powered by
# machine-learning
  • b
    Thanks that helps
  • j
    wait there is not senator armstrong
  • j
    so i put and nothing changed @User
  • c
    Send me the drive id
  • j
    why
  • c
    So I can test the model
  • j
    ok
  • j
    wait
  • j
    i think i give up because i used an personal account
  • j
    Does anyone know if there's a good learning rate number that I should put it for the spectrogram generator at TalkNet? I already tried to train with default for it and for the epochs and it doesn't sound that great.. I'm with a dataset with around 480 files.
  • j
    Also batch size.. should I let it default?
  • g
    Use 1e-4
  • c
  • j
    my first model(finaly)
  • j
    it should be damm chesse
  • j
    put i forgor that it was portuguese
  • m
    some more stonks for another character!!!
  • m
    if you can guess who it is just from the fact that i made foxy last, i give you a shiny quarter!
  • m
    okay well i finished it and well, i dont know whether this is the hifi gan or if its because this is how she normally sounds...
  • m
    here is a sample of the voice for reference:
  • m
    you know, i'll just submit her, she sounds good when speaking... maybe uberducks hifi can do something...
  • u
    sounds pretty good
  • u
    here is a test of my Funtime Freddy voice model
  • m
    i dont think you will be allowed to submit that model...
  • m
    kellen doesn't allow his voices to be cloned...
  • u
    oh ok
  • m
    still sounds cool
  • e
    I got to about 1.1 before my computer decided to crash (I don't have the exact number because of said crash) and it doesn't sound any better. It was around epoch 300
  • m
    do you mean 0.11? theres no way 300 epochs and its still 1.1
  • c
    1.1 is not good that is above 1
1...729730731...1068Latest