https://uberduck.ai/ logo
Join Discord
Powered by
# tacotron-2-support
  • r

    Reclezon

    11/05/2022, 2:26 AM
    Well now I'm lost. . . Does inputting the dataset with the link result in an error too?
  • u

    🏝✍Berkeley•Ray•Brewski

    11/05/2022, 2:29 AM
    Honestly it's my first time even trying to get a dataset; Which makes it 2% as stressful for me.
  • r

    Reclezon

    11/05/2022, 2:34 AM
    Making the actual dataset isn't hard, once you know a good setup that works for you with making them, but if the math ain't mathing on the colab, I think I'd just have to refer you to [@]hecko#8977 for that one. Afaok the colab in the nb tab should give the same results without failure?
  • u

    🏝✍Berkeley•Ray•Brewski

    11/05/2022, 2:35 AM
    Reaching out to hecko sounds like a plan! What's Afaok?
  • r

    Reclezon

    11/05/2022, 2:36 AM
    Sorry, meant
    as far as I know
  • u

    🏝✍Berkeley•Ray•Brewski

    11/05/2022, 2:36 AM
    Ah, Thanks!
  • u

    🏝✍Berkeley•Ray•Brewski

    11/05/2022, 2:37 AM
    Just in case; What are some different types of Training models that are still usable?
  • r

    Reclezon

    11/05/2022, 2:40 AM
    There's the Legacy Tacotron, which is Pipeline Lite™️, and Talknet models which Idk much about. Also HiFi-gans which are what make Tacotron models sound really good if you have a trained one.
  • u

    🏝✍Berkeley•Ray•Brewski

    11/05/2022, 2:45 AM
    HiFi-Gans sounds great!
  • m

    MapleNeko

    11/05/2022, 5:08 AM
    my old model giving this error
  • m

    MapleNeko

    11/05/2022, 5:08 AM
    i dont want to redo the training up to 500
  • m

    MapleNeko

    11/05/2022, 5:08 AM
    should i continue it?
  • h

    haru0l

    11/05/2022, 5:34 AM
    sounds like you didnt put a hifigan link
  • m

    MapleNeko

    11/05/2022, 5:38 AM
    my old model doesn't use hifigan
  • m

    MapleNeko

    11/05/2022, 5:38 AM
    is hifigan required now?
  • u

    {K EY1} (Kei)

    11/05/2022, 7:36 AM
    No But you need to have some sort of a value Leave it as the default
  • d

    Dingusmcbingus

    11/06/2022, 12:50 PM
    so my files converted to npy and now i have 1095 loss
  • d

    Dingusmcbingus

    11/06/2022, 12:51 PM
    oh wait
  • d

    Dingusmcbingus

    11/06/2022, 12:51 PM
    that was val loss
  • d

    Dingusmcbingus

    11/06/2022, 12:51 PM
    nvm
  • a

    AhmadGT

    11/07/2022, 2:26 PM
    ??
  • u

    {K EY1} (Kei)

    11/07/2022, 2:49 PM
    Sorta You want it to be straight but you also want a low enough val loss Around 0.1
  • a

    AhmadGT

    11/07/2022, 4:12 PM
    okay thanks!
  • u

    (".Lukus_bt.")

    11/10/2022, 6:53 PM
    Help please
  • u

    (".Lukus_bt.")

    11/10/2022, 7:07 PM
    ???
  • m

    MapleNeko

    11/11/2022, 12:29 AM
    "Sometimes you have to strike hard and fast at just the right moment to expose the lies while avoiding any unwanted interference." Should i train with hifi-gan now or more training needed?
  • m

    MapleNeko

    11/11/2022, 12:56 AM
    is this correct?
  • p

    postmates!!

    11/11/2022, 4:29 PM
    about to train my tord and i have 145 clips, what settings should i train woth
  • s

    Sonic2022_mario

    11/12/2022, 2:23 PM
    Sneak Peek Of A New Voice I'm Training
  • p

    PUMPKINEATER

    11/13/2022, 4:50 AM
    id normally slap your face off
1...697071...158Latest