https://uberduck.ai/ logo
Powered by
# machine-learning
  • p
    bad
  • p
    about the same
  • h
    probably should let it train for a bit longer
  • p
    probably not
  • h
    r.i.p
  • p
    link to the dataset?
  • h
  • m
    Hey everyone, feel free to DM me if you want to try out a Dall-E Demo website 🙂
  • p
    @User delete stuff like this
  • p
    delete this
  • p
    and everything that's not at least 1.5 sec long
  • h
    Alright
  • s
    @Sonikkukokoro @pi i have a possible idea for fixing the turret model/dataset if you havent already
  • s
    if the lines are too short, maybe you could combine some of them?
  • s
    instead of something like 1.wav|Crtitical error. 2.wav|Shutting down. you could do "1.wav|Critical error, shutting down."
  • n
    Copy code
    find "noice/output/$2/datasets/wavs" -type f \( -name "*wav" \) -size +680k -delete
find "noice/output/$2/datasets/wavs" -type f \( -name "*wav" \) -size -70k -delete
  • n
    bash script to do that for you automatically
  • n
    deletes wavs under 70k and over 680k
  • n
    adjust as you see fit, but those are decent numbers
  • p
    maybe
  • h
    I was thinking most likely replace the small wav files with some dialog from the movie
  • s
    I guarantee you combining short lines into longer ones would fix it
  • h
    Cause I only used dialog from the game
  • p
    The prosody then doesn't fit anymore
  • s
    wdym
  • p
    so I would bet on that
  • p
    what part dont you understand
  • s
    nvm
  • s
    ok now i see what you mean i dont think that would be an issue since the turret doesn't change its tone of voice very much, if that makes sense
  • s
    Let me try to make some examples real quick
1...219220221...1068Latest