https://uberduck.ai/ logo
Powered by
# machine-learning
  • a
    about talknet, when my colab pro is over, i'll buy one-time this.
  • u
    okay
  • d
    I have validation loss at 1249, is that bad?
  • a
    nope not rly
  • a
    most you want to pay attention is that less than 0.1
  • d
    Your validation loss is 0.109, that's good (probably), the 1249 is just the iteration number (number of epochs times number of batches, 96*13 = 1248)
  • a
    ^
  • s
    What's talknet?
  • u
    it's a different version of tacotron2 but ability to sing or emotion.
  • a
    but well, as it takes vocal audio reference for pitch
  • a
    it's fun 😂
  • a
    but it doesn't work with tacotron2 models sadly
  • a
    so ya need train for talknet sadly
  • s
    Oh hm How would you train a talknet model?
  • s
    Yeah, you need to port a dataset if you even have it.
  • a
    there's a colab for training
  • s
    As in the same dataset you'd use to train a tacotron model?
  • a
    basically you need zip dataset and list.txt to it
  • s
    This sounds interesting
  • s
    Are there any like tutorials or something
  • s
    Or just links in general how do I find this lmao
  • a
    there are links in #768215837248716819 but i'll grab em, give me a moment
  • s
    Alright cool
  • a
  • a
    there's should be synthesize at the end page
  • s
    Nice imma save that and try it out when I get home
  • s
    So I should just be able to use datasets I already have?
  • s
    I have like a bunch on my pc
  • a
    pretty much yeah
  • a
    dataset zip file and list.txt to it
1...282283284...1068Latest