https://uberduck.ai/ logo
Join Discord
Powered by
# tacotron-2-support
  • u
    Oh i feel bad for you, you retrained twice? That's not your error, that's a torchmoji error It shouldn't be a big deal unless you want to make use of torchmoji. You can still hear the output of your model. If you want to see the torchmoji output, i think the solution is to run in a new cell 
    !pip install --upgrade torchmoji
  • n
    idk what's torchmoji and the output sounds like this
  • u
    Ok i have no clue why it's doing that Bad data, probably :(
  • n
    I trained that one 2 days ago, and I didn't have any error
  • n
    no my data is very good
  • u
    Huh Can i hear a sample
  • n
    that's the output audio that I did 2 days ago
  • u
    I want to hear a sample of your input data
  • n
    very clear audio
  • n
    51 minutes dataset
  • u
    Huh... then there must be an error in training Are you sure the data was in 22050 hz 16 bit mono wav format?
  • n
    Yes
  • n
    but I did the training and it worked well 2 days ago
  • n
    but then it doesn't work anymore
  • n
    I did all steps like previously, then when I synthezise I get that error
  • u
    There must be some error in the notebook then I don't remember who made that notebook, but it should say at the top I think you should ping them
  • n
    @hecko
  • n
    @mega b
  • h
    oh that's because the emoji library updated
  • h
    hold on i have an up-to-date version
  • h
  • n
    there isn't hifigan id cell?
  • h
    -oh uhh
  • h
    i'll add that tomorrow
  • h
    for now i guess you can yoink the fix from this notebook to yours
  • h
    view code for the synthesis in the crust notebook
  • h
    and replace the last line with 
    print(emoji.emojize(f"Top emotions detected by Torchmoji: {' '.join(emojis)}", language = "alias"))
  • n
    Ok will test that rn
  • n
    yeah it still sounds terrible
  • n
    it was amazing 2 days ago, now it's shit with the same exact data
1...363738...158Latest