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

    Reclezon

    11/15/2022, 1:16 AM
    I'm sure the online Nb you'd probably have to count for ping when looking for speed vs using local. Getting results genned from your own computer would be faster by a ms or two vs from a server with the same hardware on the completely opposite side of the world from you
  • r

    Reclezon

    11/15/2022, 1:17 AM
    Asfaik I had better ping living in Georgia vs in Florida l. At least when playing on Minecraft.
  • h

    hecko

    11/15/2022, 1:27 AM
    surely that's not the main reason though
  • h

    hecko

    11/15/2022, 1:28 AM
    it might be that the notebook uses cpu like the tacotron one, not sure though
  • r

    Reclezon

    11/15/2022, 1:34 AM
    Never intended to mean it was, but it's a minor thing to note if you're dealing with heavy data. At least for me, anyway.
  • m

    MapleNeko

    11/15/2022, 2:51 AM
    this voice shouldn't be like this
  • m

    MapleNeko

    11/15/2022, 2:51 AM
    is too robotic
  • m

    MapleNeko

    11/15/2022, 2:51 AM
    does it need more training or more dataset?
  • m

    MapleNeko

    11/15/2022, 2:52 AM
    i trained it 880 epoch at default settings hifigan 220 at default settings 16 minutes 28 seconds of dataset
  • m

    MapleNeko

    11/15/2022, 2:53 AM
    compared to this yelan model i made last 5 months is much better than this model of yae
  • w

    WeegeeFan1

    11/16/2022, 1:24 AM
    Testing some things I'm still working on, and this has been persistant for months in both uberduck and the notebooks. Sometimes the auto-tuner is consistantly off on the octave it decides to autotune my voices in. It's even down to the way some words are submitted into the model.
  • w

    WeegeeFan1

    11/16/2022, 1:25 AM
    For example; "Keeps you groovy, 24 hours a day!"
  • w

    WeegeeFan1

    11/16/2022, 1:25 AM
    Correct octave above
  • w

    WeegeeFan1

    11/16/2022, 1:25 AM
    But I simply change the spelling and it screws up the octave
  • w

    WeegeeFan1

    11/16/2022, 1:26 AM
    For example here; "Keeps you groovy, twenty four hours a day!"
  • w

    WeegeeFan1

    11/16/2022, 1:26 AM
    For reference here it is without the autotune.
  • w

    WeegeeFan1

    11/16/2022, 2:16 AM
    My question is is there a way to manually override this?
  • b

    bluesky

    11/16/2022, 6:44 AM
    hi
  • u

    {K EY1} (Kei)

    11/16/2022, 1:05 PM
    There's not, it's unfortunately just a weird anomaly I find some tones of voice aren't always detected in the right octave I have that issue with a couple voices when i use Tsurumaki Maki English SynthV's voice as reference audio, but not when i use SOLARIA SynthV's voice
  • r

    Radak

    11/16/2022, 6:03 PM
    Does anybody have advice for someone who is trying to start putting their own AI’s for beginners
  • h

    hecko

    11/16/2022, 7:25 PM
    https://docs.google.com/document/d/1Gt_TNLAWyNzr0bjtYOE9Qm8yEvIL5yDq-W4E0kH5XW0/edit
  • l

    liltosh

    11/17/2022, 12:36 AM
    Dwag I’m into trafficking. What tactics should I play when I meet my client ?
  • h

    Haianh266

    11/17/2022, 8:42 AM
    hey guys, what is the true answer?
  • q

    QuboMatic2K6

    11/17/2022, 8:49 AM
    🤔
  • k

    KnightKat

    11/17/2022, 8:32 PM
    Sorry for the questions (am new here). I read the Voice synthesis: A Duckling's Guide already... Is Tacotron2 (Pipeline) just for doing TTS? And how does that compare/relate to Talknet2? I read that Talknet2 uses rhythm and pitch prediction so does that mean it can be used to emulate singing? Which one, or both are good for say having Captain Picard sing in a Michael Jackson video. Similarly, which one should I use to swap Picard into a movie? It didn't seem apparent (maybe i missed something) what the difference between them was. Finally... are these relatively easy to setup on a windows box (I just don't have a spare box to setup unix on).
  • h

    hecko

    11/17/2022, 9:13 PM
    tacotron takes in text and outputs speech, talknet takes in text and rhythm and pitch and outputs speech matching that rhythm and pitch it can be used for singing and does pretty well at it, though it usually can't do long notes
  • h

    hecko

    11/17/2022, 9:14 PM
    no idea about local setup but in principle it shouldn't be that much harder on windows than on linux (but in case it does become an issue, it's actually really easy to have both windows and linux on one machine; look up
    dualbooting
    )
  • c

    Cipher

    11/17/2022, 9:48 PM
    Copy code
    py
    def w1_tree_split_data_left(X, Y, feature_index, split_value):
        """Split the data `X` and `Y`, at the feature indexed by `feature_index`.
        If the value is less than `split_value` then return it as part of the left group.
        
        # Arguments
            X: np.array of size `(n_objects, n_in)`
            Y: np.array of size `(n_objects, 1)`
            feature_index: index of the feature to split at 
            split_value: value to split between
        # Output
            (XY_left): np.array of size `(n_objects_left, n_in + 1)`
        """
        X_left, Y_left = None, None
        XY_left = []
        for row in X, Y:
            if row[feature_index] < split_value:
                XY_left = X.append(Y)
        return XY_left
  • c

    Cipher

    11/17/2022, 9:48 PM
    Can anyone please help me, I don't understand what I do wrong here?
  • h

    hecko

    11/17/2022, 10:15 PM
    .append()
    edits the actual array itself and returns nothing i'm not sure about the syntax for numpy arrays but with python lists i'd do
    X + Y
    or
    [*X, *Y]
    though it's weird that you're setting it to the entirety of
    X
    and
    Y
    every time
1...101110121013...1068Latest