https://uberduck.ai/ logo
Powered by
# machine-learning
  • h
    (sorry my internet lagged)
  • n
    is there a way to turn those things to audio files?
  • h
    ok yeah there is exactly one | per line
  • h
    so not sure then
  • h
    if you used notepad then maybe it saved as utf-16 for whatever reason? idk
  • h
    btw all your apostrophes are gone
  • m
    it's because of this colab
  • m
    i downloaded that colab file
  • m
    uploaded it into google colab
  • m
    and first i saw that there were problems with num2words
  • m
    so i thought i would type !pip install num2words
  • h
    maybe michael uses fancy apostrophes aka ’
  • h
    personally i would just go through the dataset manually and wordify the numbers depending on how they are in the audio
  • h
    because num2words can't know whether michael pronounces it "two thousand six hundred" or "twenty six hundred"
  • m
    don't worry
  • m
    i will send you my michael mjd dataset on pm
  • m
    as a link of course
  • m
    so you can sort this problem out
  • m
    and before i do it
  • m
    i'll have to vocal isolate all of the wav files
  • h
    no thanks
  • h
    i'm probably not gonna get much computer access today
  • m
    okay
  • c
    Is this from my colab?
  • m
    yes
  • m
    idk
  • c
    I think it is
  • c
    but, if you want to convert numbers to train in tacotron 2, tacotron already does that, but I'll check the colab here to see what's the problem
  • c
    The thing is it does that for English only, that's why I made this code, so it would work with any language
  • c
    Welp, found the problem
1...633634635...1068Latest