https://uberduck.ai/ logo
Join Discord
Powered by
# machine-learning
  • d

    DJ Gummibar (mega g)

    05/09/2021, 1:20 PM
    ok
  • d

    DJ Gummibar (mega g)

    05/09/2021, 1:21 PM
    also when im done with the sunny funny dataset,i gotta upload it in #dataset
  • p

    pi

    05/09/2021, 1:23 PM
    you should definitely cut out sound effects like in this clip
  • p

    pi

    05/09/2021, 1:23 PM
    @User
  • p

    pi

    05/09/2021, 1:23 PM
    i'd also advise against the clips where he is shouting
  • p

    pi

    05/09/2021, 1:24 PM
    like here
  • d

    DJ Gummibar (mega g)

    05/09/2021, 1:43 PM
    and im almost trying to get up to 20 audios
  • d

    DJ Gummibar (mega g)

    05/09/2021, 1:43 PM
    and its gotta be a little hard to find more
  • p

    pi

    05/09/2021, 1:46 PM
    20 isn't enough
  • d

    DJ Gummibar (mega g)

    05/09/2021, 1:48 PM
    then 30
  • p

    pi

    05/09/2021, 1:48 PM
    make it 50
  • d

    DJ Gummibar (mega g)

    05/09/2021, 1:48 PM
    but i have to make it in 30
  • p

    PUMPKINEATER

    05/09/2021, 1:49 PM
    60 does good for me
  • d

    DJ Gummibar (mega g)

    05/09/2021, 1:49 PM
    its up to 30 lines
  • p

    pi

    05/09/2021, 1:49 PM
    ?
  • d

    DJ Gummibar (mega g)

    05/09/2021, 1:49 PM
    but someone just told me if 4 is enough but its not
  • p

    pi

    05/09/2021, 1:50 PM
    what?
  • d

    DJ Gummibar (mega g)

    05/09/2021, 1:50 PM
    ....
  • o

    OctolingVladTTS

    05/09/2021, 1:53 PM
    Lucy's voice, that I have finished it, have few dialogues found in the anime, and the 5 were found from the game, but I believe in the other characters they must have more dialogues or the same
  • i

    Isoar

    05/09/2021, 2:38 PM
    hi I trained my first voice a while ago and am working on my second one but I noticed at around validation loss of 0.06 the spectograms would look very different, some would have all yellow in a straight line while others looked more of a variety I wanted to know what the ideal spectogram should look like so I can get the most accuracy or if it depends on the voice and there's no telling what's the best one
  • p

    pi

    05/09/2021, 3:03 PM
    the alignment graph should look like this @User
  • i

    Isoar

    05/09/2021, 3:06 PM
    okay thanks
  • o

    OctolingVladTTS

    05/10/2021, 2:37 AM
    I'm going to try one more time, to see if it goes well, I lower the RAM to 6, in case the quality varies is the Ninjala van voice
  • o

    OctolingVladTTS

    05/10/2021, 2:37 AM
    I hope it gives me luck
  • o

    OctolingVladTTS

    05/10/2021, 2:51 AM
    What do you think?
  • o

    OctolingVladTTS

    05/10/2021, 2:51 AM
    this is what i wrote:
  • o

    OctolingVladTTS

    05/10/2021, 2:51 AM
    Here is a fun fact, I am actually taller than the Master Chief
  • w

    Waltman13 (semi-active)

    05/10/2021, 5:20 AM
    not bad.
  • u

    (Edd) bruh moment

    05/10/2021, 5:56 AM
    pretty cool
  • n

    Neotheyoshare

    05/10/2021, 7:41 AM
    Generating Mels 86% 69/80 [00:01<00:00, 37.56it/s] /content/tacotron2/utils.py:14: WavFileWarning: Chunk (non-data) not understood, skipping it. sampling_rate, data = read(full_path) --------------------------------------------------------------------------- RuntimeError Traceback (most recent call last) in () 1 if generate_mels: ----> 2 create_mels() 3 frames /content/tacotron2/stft.py in transform(self, input_data) 82 83 # similar to librosa, reflect-pad the input ---> 84 input_data = input_data.view(num_batches, 1, num_samples) 85 input_data = F.pad( 86 input_data.unsqueeze(1), RuntimeError: shape '[1, 1, 51652]' is invalid for input of size 103304
1...646566...1068Latest