# tacotron-2-support
- m
mepc36
09/16/2022, 2:09 PMyeah you're right, good call, thanks: - m
mepc36
09/16/2022, 2:13 PMHey, how does the weighting of a pretrained model with the new inputs get configured? Are there hyper parameters that do that? - u
{K EY1} (Kei)
09/16/2022, 2:13 PMYeah, the layers that are ignored are defined in hparams.py - m
mepc36
09/16/2022, 2:21 PMShit man, you're good, thank you! - u
{K EY1} (Kei)
09/16/2022, 2:22 PMThanks lol And np - o
OccultMC
09/16/2022, 11:30 PMThey’re not tortoise? - o
OccultMC
09/16/2022, 11:33 PMIf ya talking about mine lmao they aren’t tortoise lmao I wish - p
PUMPKINEATER
09/16/2022, 11:42 PMaight alright jesus - o
OccultMC
09/17/2022, 12:17 AMHuh - o
OccultMC
09/17/2022, 12:17 AMWhat did I do? - m
MrDankEngine
09/19/2022, 10:01 AMI'm trying to submit Charli's voice into Uberduck, but she hasn't popped up on the website - m
MrDankEngine
09/19/2022, 10:02 AMFor those who are wondering, Charli is from a band called "Hi-5", and if you aren't familiar with them, they're an Australian band that were part of thousands of childhoods out there, including mine - m
MrDankEngine
09/19/2022, 10:02 AMHer last name is Robinson - m
MrDankEngine
09/19/2022, 10:04 AMHey @WigglyFan2001, if you're seeing this, I hope you'll understand about my troubles of submitting her voice - m
mepc36
09/19/2022, 1:46 PMHow do you know when to stop training? - p
PixPrucer
09/19/2022, 1:47 PMwhen the validation loss starts hitting around 0.1~0.09 - p
PixPrucer
09/19/2022, 1:48 PMAt least that's how I've been taught - p
PixPrucer
09/19/2022, 1:48 PMYou want to get the loss as low as possible without overfitting the model - c
CmanflipYT
09/19/2022, 4:16 PMI would open a ticket for that. - m
MrDankEngine
09/19/2022, 4:16 PMHow do I do that? - c
CmanflipYT
09/19/2022, 4:16 PM#813143605862203442 - c
CmanflipYT
09/19/2022, 4:17 PMThen you can ask a mod in there. - m
MrDankEngine
09/19/2022, 4:21 PMI just sent them a question - u
{K EY1} (Kei)
09/20/2022, 2:09 AMgetting this when training on pipeline on the polish base model, always after iteration 99 - o
Oreo
09/20/2022, 7:17 AMHi, I've been trying to train my first voice with pipeline, and I've ran it through a couple hundred epochs. I'm not super happy with how it's sounding, but I'm not sure if it's because I haven't trained it enough, or if I didn't include enough wavs(45) in the dataset. - p
PixPrucer
09/20/2022, 7:42 AMOh it might've overfit like crazy - o
Oreo
09/20/2022, 8:11 AMShould I try retraining then? - p
PixPrucer
09/20/2022, 8:51 AMPreferably for max 100 epochs - p
PixPrucer
09/20/2022, 8:51 AMand more data if possible - p
postmates!!
09/20/2022, 10:40 AMi only have 25 or so clips in my dataset (theyre all i could find 💀💀) what settings and stuff should i use to get the best result