https://uberduck.ai/ logo
Powered by
# machine-learning
  • e
    What I was doing is I downloaded one file with all of his lines and seperated it into wavs
  • u
    just from youtube? the quality is gonna be pretty bad lol, youtube encoding isn't that great
  • e
    Yea I know. I've gotten decent results from doing that before but ur probs right lol
  • u
    this is what mine sounds like lol
  • u
    i have to spell some things phonetically to force the ai to pronounce things like i want
  • e
    def better than mine lol
  • u
    seems to have problems pronouncing 'u's correctly lmao
  • u
    i also generated this line like 5 times before i got it to not have a stroke and actually sound like i wanted it to
  • u
    still has a tiny stroke when it tries to say "I trained"
  • e
    Yea. Wheatley seems like a difficult one to do.
  • s
    i wonder why
  • a
    I have tf2 installed and i could probably bring ap-sap lines
  • a
    Ap-sap - basically sapper wheatley
  • a
    So dunno if he has same lines
  • j
    is it bad to have too many short clips in your dataset
  • j
    would it be better to join them
  • j
    and have a bunch of clips that are two sentences
  • j
    rather than a bunch of short clips
  • a
    Not really.
  • j
    ok
  • j
    thanks
  • j
    I have 243 clips but only about seven minutes
  • j
    if I want to make it better I might just have to get still more clips
  • j
  • u
    depends how short. you ideally should not have clips shorter 1.5s or longer than 12s
  • j
    they're shorter than 1.5s
  • j
    like this
  • j
    ones without a duration are shorter than one second
  • u
    you should also have a range of clip lengths, so you shouldn't have too many short and too many long clips, you want a (relatively) even spread of clip lengths
  • j
    I have a lot of short
1...271272273...1068Latest