https://uberduck.ai/ logo
Powered by
# machine-learning
  • c
    I plan on training a huge model locally, that's why I wanted to know
  • s
    yes, I believe it can, although its tricky
  • s
    try setting "distributed_run": true
  • t
    So I'm looking into whatever other methods of English and Japanese voice cloning I can, maybe I ought to keep it to focusing on developing the actual voices I want to use myself I found this https://colab.research.google.com/github/espnet/notebook/blob/master/tts_realtime_demo.ipynb?authuser=2#scrollTo=n2Dk9o0-JlbN
  • u
    Oh espnet is interesting I haven't gotten it to work yet but go for it
  • j
    Copy code
    pip install fucking
  • i
    THE FUCK
  • c
    >a python tool which make English word to Chinese using the baidu translate.
  • i
    where is the sex
  • t
    Currently delaying all plans for my bilingual model because my Shared Drive imploded and I lost the version of the dataset that was meant to be used
  • t
    Ought to focus on developing voice models for Greg and Rias, English and Japanese respectively
  • t
    Situation reminds me of that Doraemon story where Doraemon gives a Nobita a tree stump with an underground system He invites his friends and they "move in" But the next day the stump disappears with all the stuff
  • t
    Anybody else trying out this model stuff?
  • u
    I want to test crosslingual/bilingual models I'm just procrastinating on it, but i think i'll use akane and jvs002 as a quick test Akane is an english dataset, but she has a japanese accent. Ideally i'd want to use something with a native english accent but akane is my biggest english dataset
  • t
    Go ahead
  • t
    I don't mind the accentuated English for the time being
  • u
    i changed my mind
  • u
    if this ends up working, i'll have a japanese version of the english voice, and there already is a japanese version of akane so it would be a waste i'm going to use zagreus for english instead and jvs001 for japanese, since it's also a male voice
  • h
    good luck
  • u
    testing my papyrus model
  • u
    Ty, i'll need it I think it's starting to work... the first sample (epoch 10) sounded like gibberish for jvs001 but sounded sorta like the input text for zagreus
  • u
    Probably because i'm using an english base model...
  • h
    tbh jvs always had training issues for me
  • h
    even when using a japanese base model
  • u
    Oh really? I got jvs001 and 002 to work
  • u
    What issues did you have?
  • h
    i think i tried to train the one that sounds like teto
  • h
    it just barely worked
  • u
    Oh
  • h
    16 i think
1...959960961...1068Latest