https://uberduck.ai/ logo
Powered by
# machine-learning
  • u
    FUCK that reminds me i tried using jukebox to continue an old oasis demo
  • u
    i deleted all the results cause i needed drive space
  • u
  • m
    i remember the last thing i tried to make before k80's appered was rick astley singing all star, but all the results weren't that good...
  • m
    YO! SMASH ANNOUNCER STONKS????????
  • u
  • u
    damn i didnt realize i had the voice clips for the maximum impact 2 announcer just laying around
  • u
    i should try training them
  • m
    it is soooooooooooooo much better!!! it actually english!!!
  • m
    thanks to @User for helping!
  • i
    yooooo actually worked 😎 😎
  • u
    max impact 2 announcer stonks better doe 😳
  • u
    he speaks so slow tho lmao
  • j
    Feature space edits
  • m
    im testing it a little more and i think its pretty good! only problem I found is that it mispronounces words at times
  • c
    LEARNING R
  • m
    well its 100x better than current one so im gonna submit it
  • u
    i should update this with regulation A lines
  • u
    he speaks a lot more in that game
  • u
    most of the wavs i have currently are character names and even then he pronounces some of them weirdly so i think its throwing the pronunciation off
  • z
    Hacked
  • c
    discqrde ‼️
  • u
    BFDI blocky
  • u
    new iteration has better pacing
  • z
    More BFDI voices poggers
  • h
    was this with the hifi gan fine tuner colab?
  • m
    no, i just retrained the model with a better dataset
  • m
    and also i used a different model for warm starting the model
  • h
    oh with the tacotron 2 colab? that's really good! been trying to synth game voices but yeah the low number of wavs gives trouble
  • u
    low wav characters are harder but not impossible
1...683684685...1068Latest