https://uberduck.ai/ logo
Join Discord
Powered by
# talknet-support
  • j

    Justin

    02/18/2023, 12:50 AM
    dm me your dataset, lemme see what's up
  • a

    Alvise

    02/18/2023, 9:57 AM
    Hi I'm new here, I came from the uberduck written tutorial and the youtube video linked with talknet. I see that in the comments Awesome Face said that talknet isn't used anymore because of NVIDIA, what is used now?
  • g

    Gosmokeless28

    02/18/2023, 10:00 AM
    Some of the Uberduck contributors use SO-VITS-SVC Although, that architecture is not (yet) supported by uberduck
  • c

    Count JurassicJordan

    02/18/2023, 8:20 PM
    Is there any video instructions on setting up an running so-vits-svc?
  • c

    Count JurassicJordan

    02/18/2023, 8:22 PM
    Scratch that,here it is

    https://youtu.be/xQ__ZLpv4yY▾

  • c

    Count JurassicJordan

    02/18/2023, 11:11 PM
    Any info in training tho?
  • c

    Count JurassicJordan

    02/19/2023, 7:08 PM
    i figured it out...holy cow, its really good. only 2300 epochs and it sounds like a real person
  • c

    Count JurassicJordan

    02/19/2023, 11:50 PM
    So someone said talknet is being replaced with so vits right? Should I start submitting models now? Cuz I trained a MLAATR XJ9 while testing
  • u

    (Dawn) Will Draw Fictional Women

    02/19/2023, 11:50 PM
    what
  • u

    (Dawn) Will Draw Fictional Women

    02/19/2023, 11:50 PM
    no it's not
  • u

    (Dawn) Will Draw Fictional Women

    02/19/2023, 11:50 PM
    who said that
  • c

    Count JurassicJordan

    02/19/2023, 11:51 PM
    Oh wait I misread that
  • c

    Count JurassicJordan

    02/19/2023, 11:52 PM
    Sorry. Is it an architecture that’ll be added soon or should I hold off on submitting models still?
  • u

    (Dawn) Will Draw Fictional Women

    02/19/2023, 11:52 PM
    i have heard zero talk of integrating it
  • u

    (Dawn) Will Draw Fictional Women

    02/19/2023, 11:52 PM
    it COULD since its in-license unlike diff but idk
  • m

    moomoo

    02/20/2023, 6:28 PM
    can you please tell us about the info about training? whether it's local or a colab?
  • c

    Count JurassicJordan

    02/20/2023, 9:12 PM
    Sure, I’m running it locally on my system. I’d say it’s a lot easier to train so-vits than talknet and gets better results with less training time. Training takes a whole lot out of my GPU tho. If you run Linux with an nvidia card with only 12GB of vram like I do, I reccomend running it in TTY to give it every bit of your gpu. I haven’t even looked at the colab. It’s probably even easier.
  • c

    Count JurassicJordan

    02/20/2023, 9:13 PM
    I have it set up in a conda environment as well. I followed thd tutorial here, however when setting up an environment, make sure the the python version is ==3.9
  • c

    Count JurassicJordan

    02/20/2023, 9:14 PM
    I should make a video or something. I’ve learned so much while gathering datasets
  • c

    Count JurassicJordan

    02/20/2023, 9:20 PM
    I really hope the architecture for this gets added to uberduck cuz it’s, imo, the best for voice refrence conversion, and sounds amazing with singing
  • m

    moomoo

    02/20/2023, 9:20 PM
    thanks for the details
  • m

    moomoo

    02/20/2023, 9:27 PM
    can I ask about how to make the dataset and the format of the audios
  • c

    Count JurassicJordan

    02/20/2023, 9:28 PM
    You can just make it any wav file at any sample rate, you put them in the dataset_raw folder and run the resample.py and it will automatically process and sample the files to the required sample rate
  • m

    moomoo

    02/20/2023, 9:31 PM
    okay thank you so much
  • c

    Count JurassicJordan

    02/20/2023, 10:12 PM
    No problem ^~^
  • m

    Makaveli AI

    02/27/2023, 1:55 PM
    anyone getting problems with training the hifigan step 11
  • m

    Makaveli AI

    02/27/2023, 1:55 PM
    it stops with this error before doing the epoches was fine 24 hours ago
  • e

    EBoyd

    03/01/2023, 9:58 PM
    Is anybody else having issues with the controllable? I can’t get it to work.
  • r

    RubbishNubbish

    03/02/2023, 6:31 AM
    has anyone seen this error
  • r

    RubbishNubbish

    03/02/2023, 6:31 AM
    ImportError: cannot import name 'get_current_traceback' from 'werkzeug.debug.tbtools' (/usr/local/lib/python3.8/dist-packages/werkzeug/debug/tbtools.py)
1...636465...74Latest