https://uberduck.ai/ logo
Powered by
# machine-learning
  • u
    i mean thats what we do on the site already
  • u
    every model is on the alex trebek hifigan by default
  • u
    yes
  • u
    train a model on an arpa dataset
  • u
    then synthesize with pronunciation dictionary on
  • h
    arpa
  • u
    @User epic mega phishing link filter fail
  • s
    oh it caught it... it's just that they sent it right afterwards lol
  • u
    i repeat
  • u
    epic mega phishing link filter fail
  • p
    How can you "debug pitch" in the talknet synthesis colab, my synthesis sounded good yesterday and now it sounds worse.
  • p
    nvm
  • u
    Is this where I go for help on using tacotron?
  • u
    Im on the training notebook section
  • y
    Usually someone will help out in here if you ask the right questions. That guide on the training notebook is extremely straightforward though
  • u
    I gave up lol
  • u
    I'm trying again tomorrow morning
  • d
    Is there such thing as too many datasets?
  • d
    Is 2000 wav files overkill?
  • u
    probably yeah
  • u
    unless youre prepared for hours of training
  • d
    Hours meaning in the dozens?
  • u
    maybe
  • d
    How many files do you suggest?
  • u
    training just 560 took quite a while for me so i cant imagine 2000
  • u
    uhhh
  • u
    60 bare minimum, 120 decent amount
  • d
    Did you use google collab or your own gpu?
  • u
    colab
  • u
    local training takes longer in many cases
1...708709710...1068Latest