https://uberduck.ai/ logo
Powered by
# machine-learning
  • u
    epoch amount i find is completely arbitrary
  • p
    How much data do you have
  • k
    windows ten
  • m
    do you have miniconda?
  • j
    269 wav files, ~11 minutes worth
  • k
    I don't think so no
  • u
    i just set epoch amount to a ridiculous number and manually stop training when i get a good val loss
  • h
    i already linked this earlier but they want to use grid3 which uses sapi which there is no tacotron wrapper for afaik
  • u
    tacotron is all about val loss ive found
  • k
    just need voice then
  • k
    I installed gui
  • k
    ty again
  • k
    I guess I just need to ask zwf
  • k
    about ccrossbreed
  • j
    So lower val loss is better? Is there a sort of ideal range I should shoot for?
  • k
    @User right?
  • k
    ooops
  • p
    No
  • k
    ddidn't mean to tag
  • p
    Just do it by ear
  • p
    Train, check
  • u
    for a dataset of your size id recommend sub-0.095
  • u
    dont go too low though or you might experience some pronunciation/tonality weirdness
  • k
    sorry for the weird questions
  • p
    I don't look at the loss
  • u
  • p
    It's okay!
  • k
    and ty mega
  • p
    Oi atleast I can use Talknet cuz my dataset be CLEAN brother
  • m
    are you still interested in getting GPU inferencing working?
1...774775776...1068Latest