https://uberduck.ai/ logo
Powered by
# machine-learning
  • h
    How
  • e
  • e
    It’s that easy 😄
  • e
  • h
  • h
    Idk I don't know how to fix this crust model, I am so close but I am doing something wrong clearly
  • u
    if this is crust hifi warm starting its normal
  • h
    So do I let it get to 6000 epochs and not steps?
  • u
    get to 6000 steps
  • u
    minus where it starts at
  • s
    Ok?
  • h
    I don't even remember where it started I am just gonna cross my fingers an pray it started at 2520000,
  • e
    You gonna be praying for a long time.
  • h
    better then praying for nothing I guess
  • h
    it's at 2524710 atm
  • h
    it's not saving the hifigan
  • h
    keeps going back to 2524510 steps and 113 epochs
  • h
    I am gonna try to add some more samples too the model
  • h
    I am worried because last time I added these I did not get good results due to the quality
  • h
    hopefully EQ matching them will be enough
  • h
    Their is a plethora of samples I could used but they are covered by another voice. I have tried Melodyne to remove a lot of it and I have gotten pretty close but unfortunately it does take a hit to the quality specularly and even worse is there voices are similar enough were they are not easily separated
  • h
    If anyone knows of really good voice splitters let me know
  • h
    I could probably try Hybrid Demucs Music Source Separation though idk how well it would do splitting voices of that quality
  • h
    It might even just bunch the two together
  • h
    I remember reading something somewhere about a project with ai that dealt with overlapping voices but I can't remember what it was
  • h
    Or if it was easy to use
  • u
    Blocky say me!
  • h
    How big is your dataset?
  • u
    Hahahahha that's such a silly answer, zop zap? Hahahha
  • t
    I’m trying to use Google Colab to try the blacklisted Numbuh 2 model but nothing’s working?
1...911912913...1068Latest