https://uberduck.ai/ logo
Join Discord
Powered by
# talknet-support
  • m
    Ty, I tried that tho and it didnt work
  • m
    It's probably because I made the model a long time ago and it's not a zip
  • m
    ig i'm going to make a new model and try again , thanks (:
  • w
    You should make it a zip
  • w
    Your welcome for whatever I taught you
  • s
    hmm guys, can 10 too long samples out of 445 mess up a model this bad
  • c
    No, this is an error with the hifigan training
  • c
    What you can do is change the generator inside of the talknet model to any other so you can still use your model
  • c
    But it won't sound that good
  • m
    i got this another problem while setup
  • m
    does anybody know why on the controllable talknet colab when I upload a reference audio then generate the audio it just, doesnt work? like nothing get generated
  • m
    it only works when I dont put any reference
  • w
    How do people make good voices that can sustain notes well like the SpongeBob one?
  • w
    Wait I'll show what I mean
  • w
    For example, Chandelier (as sang by Puddles Pity Party)
  • w
    That's my source audio
  • w
    Now I'll show what the SpongeBob voice on Uberduck.ai sounds like
  • w
    This is my Cab voice:
  • w
    With autotune and the "reduce matallic noise (slow) option in the notebook:
  • w
    Without autotune nor matallic noise filtering:
  • w
    See it's not great
  • w
    How exacly do I make it better? Is it a training thing or a dataset issue?
  • w
    I've got clips of him sustaining notes in my dataset
  • w
    Such as:
  • w
    You see I really don't understand it
  • w
    Can I just get some advice please
  • w
    E
  • s
    I'm getting the same error
  • m
    why did this happen ?
  • i
    I can't figure out what val_filelist it is.I wanna train talknet but i dont know what val_filelist in the step 3
1...383940...74Latest