https://uberduck.ai/ logo
Join Discord
Powered by
# talknet-support
  • w

    WeegeeFan1

    11/10/2022, 4:38 PM
    The val filelist is just a thing to make sure the AI isn't regurgitating what it's learning and actaully doing original things.
  • w

    WeegeeFan1

    11/10/2022, 4:39 PM
    You need to take a third of your transcriptions from your main transcriptions text file and save it into another file, removing them from the first one
  • w

    WeegeeFan1

    11/10/2022, 4:39 PM
    Then you upload that
  • w

    WeegeeFan1

    11/10/2022, 4:39 PM
    to the field where it asks for a validation thing
  • s

    Sonic2022_mario

    11/13/2022, 8:26 PM
    Someone Test This With A Song https://drive.google.com/file/d/17mVUcbc8Dfed1znmBucH4odRotnntYcd/view?usp=share_link
  • w

    WeegeeFan1

    11/14/2022, 9:50 PM
    I'll do that. Don't know what the voice is. I'll make him sing 30s jazz because why not.
  • w

    WeegeeFan1

    11/14/2022, 9:52 PM
    oo it's kinda scary
  • w

    WeegeeFan1

    11/14/2022, 10:30 PM
    I can't really autotune it because it's a crowd, but they just come back terrible anyway.. I'm not sure what song they'd really fit in..
  • s

    Sonic2022_mario

    11/14/2022, 10:40 PM
    Try Without
  • w

    WeegeeFan1

    11/15/2022, 1:34 AM
    For what song? I literally can't think of any that it would fit for.
  • w

    WeegeeFan1

    11/15/2022, 1:34 AM
    Nevermind i found a song

    https://www.youtube.com/watch?v=kbHfvvk9Q5s▾

  • m

    Mann Maker Studio

    11/15/2022, 10:23 AM
    is there a tutorial video of how to train a model in tacotron2 pipeline ?
  • s

    Sonic2022_mario

    11/15/2022, 12:24 PM
    Now To Test
  • u

    {K EY1} (Kei)

    11/16/2022, 1:08 PM
    No, but try to follow the directions on the cells If you're confused, ask a specific question in #994486394049282058
  • p

    postmates!!

    11/16/2022, 9:41 PM
    for a 145 wav dataset should i train with learning decay?
  • g

    Gosmokeless28

    11/17/2022, 2:09 AM
    Yeah
  • p

    postmates!!

    11/17/2022, 2:57 AM
    meant to message in tacotron woops
  • w

    WeegeeFan1

    11/19/2022, 12:05 AM
    I wonder if the reason this channel is so quiet is because nobody needs help and everyone understands it or because barely anybody actually makes any models..
  • s

    Sonic2022_mario

    11/19/2022, 1:35 AM
    Gonna Test My Singing Kids Model?
  • w

    WeegeeFan1

    11/19/2022, 1:35 AM
    OH I forgot! I'll do it
  • s

    Sonic2022_mario

    11/19/2022, 1:36 AM
    Send Results When Done
  • w

    WeegeeFan1

    11/19/2022, 1:37 AM
    Yes sure I will. I think it'll be kinda werid since the song is meant to be sang by an individual singer. Not a creepy crowd of kids
  • w

    WeegeeFan1

    11/19/2022, 1:37 AM
    Unless that's the point
  • w

    WeegeeFan1

    11/19/2022, 1:38 AM
    Should I auto-tune it into pitch or just leave it as untuned? It sounds better untunted but it's off-pitch sometimes
  • s

    Sonic2022_mario

    11/19/2022, 1:38 AM
    Auto tune it
  • w

    WeegeeFan1

    11/19/2022, 1:40 AM
    Auto-tuning it doesn't really sound natural especially since a crowd of kids wouldn't be perfectly in pitch anyway. And the auto-tuning treats it like a single voice, as it should.
  • w

    WeegeeFan1

    11/19/2022, 1:40 AM
    But with your crowd it doesn't work correctly
  • w

    WeegeeFan1

    11/19/2022, 1:40 AM
    I'll show you a comparison in a few minutes when i have something
  • w

    WeegeeFan1

    11/19/2022, 1:41 AM
    You like the song I'm doing?
  • s

    Sonic2022_mario

    11/19/2022, 1:41 AM
    Yep
1...394041...74Latest