https://uberduck.ai/ logo
Powered by
# machine-learning
  • h
    I tried converting them to mono this time but that still happens
  • j
    Nah that's not the issue
  • j
    The problem is the audio clips have too much background noise
  • j
    Lemme share an example real quick
  • h
    Can you fix them for me again? Just do clips 21-40 because those are new
  • j
    Here's an example
  • j
    I'll see
  • h
    ok
  • h
    Are you going to do it?
  • j
    Kinda busy atm
  • j
    so can't guarantee
  • h
    I can't do this I'm sorry
  • h
    I tried using the vocal remover 5 notebook but it's like complicated
  • u
    vince has turned into black and white
  • u
    also new nickname
  • m
    Vince being Uncanny
  • c
    congratulations you have entered triple digits
  • c
    your dataset is now: GOOd
  • c
  • j
  • c
    he is FINALLY GETTING ALIGNMENT IMAGE
  • u

    https://media.discordapp.net/attachments/364889805898776576/894378871330320414/caption.gif▾

  • p
    So uh, is Tacotron the only neural network out there that can synthesize audio? Because like...iirc openai jukebox is also based on tacotron, but it synthesizes full musical tracks...
  • p
    https://github.com/ninon-io/Impact-Synth-Hardware/ but i just found this and it looks like they're using another engine
  • p
    just labeled as "Pytorch deep audio synthesis model"
  • c
    there are other networks that predict mel spectrograms like tacotron (see: speech synthesis section in NVIDIA/NeMo )
  • c
    since they predict mel spectrograms they are often paired with a vocoder like hifigan
  • p
    ahhhhh
  • p
    so by "vocoder" you just mean like, a FFT resynthesis engine?
  • c
    i guess
1...678679680...1068Latest