https://uberduck.ai/ logo
Powered by
# machine-learning
  • m
    don't use ... cause it doesn't work well
  • t
    How about this "Oh no the tornado is gonn-" Sonic: "Sorry to interrupt you tails" it’s when someone could not finish the sentence cause someone else interrupted him.
  • t
    The sonic message is just a example
  • m
    i would reccomend not using the - and replace it with a full stop
  • h
    i'd actually recommend using the 
    -
    so it doesn't learn to stop 
    .
    sentences abruptly like that
  • m
    https://colab.research.google.com/drive/12p0jpSx3pJNfZk-y_L44yeHZlhsKVra-?usp=sharing MTTR - End to End Referring Video Object Segmentation AI Detects objects AI and tells what object is or so I called end of hard rotoscoping
  • j
    Seems cool but its not as reliable as rotobrush 2.0 or runwayml
  • w
    a bit random but has anyone tried training the smash 64 and melee announcers with an arpabet-based transcript?
  • a
    did anyone ever make a model for the characters in Final Fantasy 7 remake?
  • a
    I just ripped all the audio files from the game, but figure I should ask here first to see if it's been done yet
  • a
    since the game is like a few years old
  • u
    if its not on the site its a safe bet the answer is no
  • u
    its not a request
  • a
    hmm, freaken like 80 gigs of wav files
  • f
    Still
  • a
    that's a pretty rude response if that's directed towards me
  • m
    I’m working on a smash 64 model but I paused to work on other stuff, but I might try it with arpabet if smash ultimate works good with it!
  • j
    I was considering doing a pitched down version of Cloud's voice, but went with downpitched Viggo Mortensen instead, so...afaik the answer to this question is no
  • j
    Great, it's whining again. Gonna try uploading the files loose and see if that does it.
  • j
    That solved it, though I'm not sure why it was complaining. Is it somehow not able to make .npy out of the .wavs if they're zipped up? Cause that's really all I can think of...
  • j
    ...is there a typo in the notebook code? Cause I'm pretty sure there's no reason for it to be looking for .wave
  • u
    theres a typo in the transcript
  • u
    it gets the file directories from the transcript
  • j
    Yep, just found it. I'm kinda surprised there was only one typo tbh.
  • j
    Disconnected after 139 epochs. Feelsbadman
  • a
    i know i'm late but
  • a
    the problem's "255.wave"
  • a
    remove the e in ".wav" and it'll work
  • j
    Already done. Unfortunately, now it's not letting me connect to a gpu so I can't start training the model again.
  • j
    Is there any way I can get a better result out of this? I keep getting kicked from the gpu before I reach 250 epochs, but even when I manage to get closer, the quality doesn't seem to get any better.
1...765766767...1068Latest