https://uberduck.ai/ logo
Join Discord
Powered by
# machine-learning
  • j

    Justin

    10/20/2022, 6:21 PM
    lol takedown moment
  • j

    Justin

    10/20/2022, 6:22 PM
    https://tenor.com/view/violin-tiny-cute-gif-15898508
  • u

    (Dawn) Will Draw Fictional Women

    10/20/2022, 6:25 PM
    what if it stayed up that would be funny i think
  • h

    hecko

    10/20/2022, 6:43 PM
    there's no doubt a torrent of it
  • j

    Justin

    10/20/2022, 6:49 PM
    I saved boffum on my gdrive
  • j

    Justin

    10/20/2022, 11:48 PM
    the flag has been removed now lol
  • j

    Justin

    10/20/2022, 11:48 PM
    it's official now
  • j

    Justin

    10/20/2022, 11:48 PM
    Turns out it was all a big confusion
  • j

    Justin

    10/20/2022, 11:49 PM

    https://youtu.be/ct3hJ8z9e8Q▾

  • u

    🏝✍Berkeley•Ray•Brewski

    10/20/2022, 11:50 PM
    https://tenor.com/view/cr1tikal-penguinz0-meme-woo-yeah-baby-gif-19316511
  • h

    hecko

    10/21/2022, 12:11 AM
    but is it actually better
  • u

    {K EY1} (Kei)

    10/21/2022, 4:19 PM
    omg... this is just the first step of the training and it's gonna take an hour this is for my talknet pretrained model btw.
  • p

    PixPrucer

    10/21/2022, 5:03 PM
    I didn't know you could pre-train talknet
  • u

    {K EY1} (Kei)

    10/21/2022, 5:26 PM
    yeah the notebook uses a pretrained model
  • u

    {K EY1} (Kei)

    10/21/2022, 5:27 PM
    which is presumably just ljspeech because it doesn't work very well with male voices and non-american accents
  • c

    Couch

    10/21/2022, 5:27 PM
    british base model
  • u

    {K EY1} (Kei)

    10/21/2022, 5:27 PM
    also i had to reset the runtime because it turns out a100s arent supported so i lost like an hour of my time there
  • u

    {K EY1} (Kei)

    10/21/2022, 5:27 PM
    the existing talknet base model is british?
  • u

    {K EY1} (Kei)

    10/21/2022, 5:27 PM
    or is that not what you meant
  • u

    {K EY1} (Kei)

    10/21/2022, 5:29 PM
    if you're wondering if my pretrained model has british data in it, it does
  • u

    {K EY1} (Kei)

    10/21/2022, 5:35 PM
    anyways back to square one...
  • h

    hecko

    10/21/2022, 5:46 PM
    talknet can't work with a british accent because arpabet is inherently american
  • p

    PixPrucer

    10/21/2022, 6:20 PM
    Enough data and it'll cope 💥
  • u

    {K EY1} (Kei)

    10/21/2022, 6:29 PM
    well arpabet tacotron works with it so
  • u

    {K EY1} (Kei)

    10/21/2022, 6:34 PM
    Dataset loaded with 46864 files totalling 55.97 hours
    this is how many hours my talknet pretrained model has!
  • h

    hecko

    10/21/2022, 6:41 PM
    holy hell
  • h

    hecko

    10/21/2022, 6:42 PM
    not optimally though
  • h

    hecko

    10/21/2022, 6:42 PM
    plus tacotron is more flexible by nature
  • h

    hecko

    10/21/2022, 6:42 PM
    whereas talknet is fused to the handrails that are the duration and pitch inputs
  • u

    {K EY1} (Kei)

    10/21/2022, 6:44 PM
    The arpabet will be assigned to phonetics regardless, and it will be consistently assigned because of the dictionary I don't see how british english would work any less effectively than american. It's just a change in some phonetics
1...996997998...1068Latest