https://uberduck.ai/ logo
Join Discord
Powered by
# tacotron-2-support
  • m
    Anyone got a link to a command/github repo/Colab notebook for that?
  • m
    All the ones I try transfer timbre too :/
  • t
    synthesizer broken again?
  • j
    Can the Tacotron notebook work on CPU? I have a big server at work with no GPU I can use for days
  • h
    synthesis can yeah, in fact it runs on cpu only by default
  • h
    training can theoretically work on cpu but please don't
  • h
    it's much more effective to just do it on the free colab quota
  • j
    google colab scammy ok ill give up
  • u
    You can run colab without credits though It's not that bad
  • j
    It kicked me off before I got to 120 epoch they are mean
  • u
    Well you can always resume later, it's not a big deal
  • j
    I want to do it now
  • u
    You know that you can't get everything for free
  • r
    I think I can get 200 epochs with datasets that are ≈100 wavs or less on the free version, given that mothing else goes wrong like a bad connection
  • j
    just something shady going on with their pricing i used to pay the 60 a month or whatever
  • j
    idk sml
  • t
    Is there a synthesis notebook that works with pre-trained models? Because https://colab.research.google.com/github/justinjohn0306/FakeYou-Tacotron2-Notebook/blob/main/FakeYou_Tacotron_2_Training.ipynb doesn't work with any of mine.
  • j
    This nb is for training a model
  • t
    I meant to send the link of the synthesis one
  • j
    Synthesis (CPU): https://colab.research.google.com/github/justinjohn0306/FakeYou-Tacotron2-Notebook/blob/main/FakeYou_Tacotron2_Hi_Fi_GAN_(CPU).ipynb
  • j
  • t
    Yeah, I've tried this one with my models, and it goes through, but nothing audible. Just some loud noise comes through.
  • j
    Oh these don't work for you anymore?
  • j
    Try enabling ``pronounciation_dictionary``
  • j
    If it still doesn't work, then your model is broken
  • t
    Ok, im about to try that
  • t
    Ok, that worked. Thanks so much 😄
  • j
    np
  • s
    Doing Dorkly Metal Sonic
  • d
    I trained my dataset to make a model but when I tested it out it just sounds like this:
1...818283...158Latest