# machine-learning
- h
hecko
09/12/2021, 8:48 PMit turns the spectrogram output by tacotron into actual sound finetuning it won't fix pronunciation or intonation but it might make it sound more realistic - u
(Dawn) Will Draw Fictional Women
09/12/2021, 8:48 PMhmm - u
(Dawn) Will Draw Fictional Women
09/12/2021, 8:48 PMi think the model handles pronunciation well enough - u
user
09/12/2021, 8:48 PMhmg - u
(Dawn) Will Draw Fictional Women
09/12/2021, 8:49 PMi'll run it through that sometime and see what it does - u
user
09/12/2021, 9:25 PMScott the Woz (2021 Edition) voice test "Uberduck is a tad bit pog" - u
user
09/12/2021, 9:26 PM - u
user
09/12/2021, 9:26 PMI'm still training it so - u
user
09/12/2021, 9:27 PM180 voice clips - u
user
09/12/2021, 9:31 PMSounding really good ngl - r
RetroGameFan
09/12/2021, 9:43 PMStonks - n
n1cte
09/12/2021, 9:45 PMDude what learning rate did you use, and also what batch size???? - u
user
09/12/2021, 9:49 PMwell idk what the learning rate is, I assume you mean Epochs so it's 169 and my batch size was 17 - n
n1cte
09/12/2021, 9:52 PMOh so you only changed the batch size oh ok 👍 - d
Dimitri Rascalov
09/12/2021, 9:52 PMnice job - u
user
09/12/2021, 9:54 PMHoping to drop this if training goes well - c
cengi
09/13/2021, 12:01 AMI just wasted an hour recording my voice and training it. No use for the model whatsoever lmao - c
cengi
09/13/2021, 12:01 AMworked though - u
user
09/13/2021, 12:02 AMDropped the Scott the Woz 2021 voice model - u
user
09/13/2021, 12:02 AMgive contributer role now - u
user
09/13/2021, 12:14 AMHeya! I wanted to get into creating voice models and stuff. Can anyone give me a quick rundown how I should go about it? - z
Zephyr
09/13/2021, 12:21 AMPick a favorite character, or a character from something obscure, collect a bunch of voice clips, transcribe them with in a one text file, and then train them using the uberduck tacotron2 training colab. - z
Zephyr
09/13/2021, 12:22 AMhttps://awesomeface349.gitbook.io/uberduck-written-tutorial/ If you want a more in-depth tutorial. - u
user
09/13/2021, 12:22 AMThanks! - c
cengi
09/13/2021, 12:31 AMIf i want to train an artist ai. I need talknet right ? - z
Zephyr
09/13/2021, 12:37 AMYou don't have too, but it's recommended to do so. - u
user
09/13/2021, 12:37 AMWhat's Talknet? - z
Zephyr
09/13/2021, 12:39 AMThat's for making characters sing/ mimic a reference recording. Regular models don't sing on their own and need to be retrained for Talknet. - u
user
09/13/2021, 12:40 AMWell then I'll make sure to do that