https://uberduck.ai/ logo
Join Discord
Powered by
# machine-learning
  • i
    english
  • i
    i try to change language but i still gets error
  • i
    If i change symbol set to default,nvidia_taco2.Was given this error.
  • h
    oh that'd be because you need to give the path to the unpacked dataset, not the google drive id
  • h
    it's unintuitive i'll admit
  • i
    how long does training the HiFi-GAN?
  • i
    I just never dared to train HiFi-GAN
  • h
    like 3 hours roughly
  • i
    ok thanks
  • j
  • c
    Is it possible to fine-tune the model?
  • j
    nope
  • i
  • m
    well, when i start up tortoise tts, it crashes
  • z
    Reminder to folks that the NVIDIA speech seminar is today
  • e
    It's truly crazy, this era of AI we're living in
  • e
    I wonder if there's good models yet that tries to turn/upscale low quality/low bitrate audio into better quality
  • e
    Because playing around with frequencies of the audio in a program usually or something like that doesn't really help
  • e
    I would love to hear old games' and series' compressed audio in HQ
  • r
    Wasn't there a video about that for sm64?
  • e
    For SM64 they actually know the source instruments that were used to create the songs so you can just get those and re-sample those to make the song in higher quality
  • u
    There are I dont know of any off the top of my head though
  • r
    If an image generator can't even semantically label its own images it produces with at least 75% accuracy, is it even a good model?
  • r
    I say no
  • h
    if you mean stable diffusion turning e.g. body parts into other body parts during img2img, you're judging a fish by its ability to walk
  • h
    or rather you're judging einstein by his ability to do a flip
  • h
    it wasn't trained to convert images to images it was trained to convert noise that it can sorta convert from images at all is just a neat bonus
  • a
    I was trying to create Pony Head's voice (from Star vs The Forces of Evil) but I don't know if the voice was good, what is your opinion?
  • z
    It's just LJSpeech
  • z
    You trained the voice poorly
1...99910001001...1068Latest