# machine-learning
- j
jens
03/14/2022, 3:53 PMi think its like at 16 hours training rn - j
Justin
03/14/2022, 4:41 PMDFL - j
jens
03/14/2022, 4:41 PMYep - j
Justin
03/14/2022, 4:42 PMAre you pre-training the network? - j
jens
03/14/2022, 4:42 PMIm currently training on iperovs pretrained rtm model and faceset - j
Justin
03/14/2022, 4:42 PMAh nice! - j
jens
03/14/2022, 4:42 PMits lower res / dims but it trains like crazy - j
Justin
03/14/2022, 4:43 PMWhat's the model arch? - j
jens
03/14/2022, 4:44 PMliae-udt, this was like only 3-4k iterations - j
jens
03/14/2022, 4:44 PMand this is 104k - j
Justin
03/14/2022, 4:45 PMAre you training this locally? - j
Justin
03/14/2022, 4:45 PMOr using Google Colab? - j
jens
03/14/2022, 4:45 PMWell im training on colab because I still want to game etc, recently got colab pro to train faster - j
jens
03/14/2022, 4:46 PMIts about as fast as my 3060 now - j
jens
03/14/2022, 4:46 PM(p100) - j
Justin
03/14/2022, 4:46 PMYe - f
F.B
03/14/2022, 4:47 PMAmite getting constant stonk so satisfying - j
jens
03/14/2022, 4:47 PMNice - u
⫩ DanielDaBloxGamer ⫩
03/14/2022, 5:47 PMI am going to use Audacity and Tacotron 2 on my own voice - h
hecko
03/14/2022, 6:38 PMooh may i recommend shtooka recorder it's janky but it splits the sentences for you, and from there it's possible to convert them into a dataset with a script - h
hecko
03/14/2022, 6:39 PMor you can sign up for mozilla common voice and record sentences there and download them upside being that you'd be contributing to their speech recognition dataset while also getting your own downside being that they only have single sentences which aren't ideal - u
(Dawn) Will Draw Fictional Women
03/14/2022, 8:11 PMI HAVE AWESOME GRAPHS!! - t
The Wiggles and High School D×D
03/14/2022, 8:17 PMDownloaded the Japanese captions for DxD New - t
The Wiggles and High School D×D
03/14/2022, 8:17 PMNow I need to figure out who says what - t
The Wiggles and High School D×D
03/14/2022, 8:17 PMSo I can use that data to hopefully start building a Japanese voice - t
The Wiggles and High School D×D
03/14/2022, 8:22 PMI cannot think of a place where I could possibly discuss that kind of work - h
hecko
03/14/2022, 8:28 PMusually i'd suggest #842883733195456572 but you're unfortunately not a contributor so i guess this will have to do or you can ask for a language channel to go alongside #903712295534882897 and such - g
grkb
03/14/2022, 11:03 PMdon't know if this is the right place, but what's up with the fact that my model(?) only works with numbers, not words? - g
grkb
03/14/2022, 11:05 PMlike i can output something like "1984" and it works but any words and it just outputs... weird sounds