https://uberduck.ai/ logo
Powered by
# machine-learning
  • s
    hmm, doesn't sound too different at 100
  • s
    alright time to reach 150
  • s
    175, still sounds kinda robotic
  • s
    going to 200
  • s
    😬
  • s
    that sounds bad
  • s
    oh wait
  • s
  • s
    yeesh
  • s
    good alignment but mixed results
  • s
    i'll stop at 200
  • s
    hmm, hoping i didnt overfit
  • s
    FIX YOUR SHIT DISCORD
  • s
  • s
    it'll probably sound better with The Juice™️
  • s
    🤔
  • s
    guess there's just one way to truly find out
  • u
    I trained five and look at this
  • s
    sounds pretty darn Cleveland
  • s
    yea
  • s
    lucky
  • e
    Well the wheatly model is going very very well =']
  • e
    Could anyone give me some pointers on what could be wrong with my dataset? Here is the training chart.
  • e
    It sounds very bad as you heard here
  • e
    I have 100 clips
  • u
    You're working on a wheatley model too? I have a medium-sized dataset (380 clips, 21 mins of audio), although it was not too great (way too many short clips, and too many emotions) so I'm probably gonna make another one and try again
  • u
    wheatley is very expressive, and he has probably the most broad range of emotions in the entire game, so a large dataset is gonna be needed before we get something that sounds good
  • e
    I don't have many short clips, but a lot of them have shouting plus I only have 100
  • e
    I didn't expect much out of my dataset lol
  • u
    yeah, i ripped all the game audio and there's like 800 wavs with about an hour of audio data, although at least half of that is unsuitable for a dataset
1...270271272...1068Latest