https://uberduck.ai/ logo
Join Discord
Powered by
# talknet-support
  • d
    Is there any talknet tutorials? I tried looking on YouTube and found nothing
  • w

    https://www.youtube.com/watch?v=aAf8HSeU9wk▾

    that's a good one and there's another video linked in the description that helps with preparing the lines
  • l
    I Submitted my Talknet gummy sons voice on UberDuck And it should be on the category called "Jokes" and the command should be "/tts_GummySons" and the tags should be "gummy sons"
  • l
    open a ticket if you want the model to be submitted
  • l
    how do i open a ticket???
  • l
    and where do i do that??
  • u
    #813143605862203442
  • l
    oh ok
  • l
    okay i did open a ticket
  • l
    what should i do next???
  • m
    models don't instantly get submitted you know...
  • l
    oh
  • l
    ok
  • d
    If you already have a Tacotron dataset, is there anything you need to change in the dataset to make it a talknet dataset or is it the same?
  • p
    Its the same.
  • d
    rum death they should mean the same to you
  • u
    it's the same, but talknet doesn't support arpabet transcriptions
  • u
    which means try not to use weird words
  • u
    that automatic arpa would get wrong
  • c
    Am I not the only one that couldn't get talknet to synthesize? It's just loading infinitely.
  • c
    Nvm, it's working
  • p
    nvm fixed it
  • p
    when should i stop training it? i know theres some way to know but i forgot
  • u
    Talknet just run each step until it's done, except for the hifigan step which you can stop after 2000 steps
  • g
    Yes
  • c
    In my tests I found out lower batch sizes works best with less data, is that true?
  • u
    Seems true to me
  • m
    Hi
  • m
    What is a good val_loss on talknet?
  • u
    For the stage where you can edit the epochs, train for 200 first and add 100 epochs each time until you see 'best loss not reached'
1...101112...74Latest