https://uberduck.ai/ logo
Join Discord
Powered by
# machine-learning
  • w
    I'll go do that.
  • j
  • w
    Those are really cool
  • w
    I generally don't use the word to image stuff nor do I use the depth perception AIs as I have no need for them, thouhg
  • w
    And I don't really think I need to get into a new type of AI just to make a picture for another AI
  • w
    But I will put them on the list of stuff to fiddle with when I have time
  • m
  • u
    Sending help in a private message.
  • g
    @sity wrong channel
  • a
    hey, does anyone know if there's an issue with the talknet collab? For the HifiGAN part, it's giving the error "IsADirectoryError: [Errno 21] Is a directory: '../wavs/'" when it's trying to do the train.py part
  • g
    #825016537949011969
  • a
    omg, the day I come back is the same day the problem starts. lmao
  • a
    alright, thanks for the heads up
  • g
    no prob
  • a
    by the way, the whisper collab is amazing. I saw a random youtube video a few days ago for whisper AI and was like "man, I wonder if uberuduck already has a collab set up for that yet". The transcription was always the most annoying part. That and the wav cut ups.
  • r
  • w
    Is saving fixed yet?
  • i
    I try to make 10 years girl voice in russian.its not to good but no to bad.
  • a
    Is it training good?
  • a
    I edited the epoch to 500
  • a
    and I put the batch size to 4
  • u
    Probably good
  • u
    Val loss is really high though
  • u
    So probably not hm
  • a
    I edited the epoch to 20, because I uploaded just 2 wavs audios
  • a
    and I put the size to 4 and the learning rate to 1e-4
  • u
    Batch size should be lower than the amount of files Set it to 1
  • a
    Okay
  • a
    That was the result
  • a
    Now I'll test the voice
1...101410151016...1068Latest