# tacotron-2-support
- t
tylerdurdenceketi
12/12/2022, 10:48 PMForget both of them. Keep an eye on inference graph. - a
AhmadGT
12/12/2022, 10:50 PMthis one? - t
tylerdurdenceketi
12/12/2022, 10:50 PMLooks perfect. - a
AhmadGT
12/12/2022, 10:50 PMreally - t
tylerdurdenceketi
12/12/2022, 10:51 PMYour validation loss is great - t
tylerdurdenceketi
12/12/2022, 10:51 PMWhich is 0.10 - t
tylerdurdenceketi
12/12/2022, 10:51 PMDid you use Aitch? - a
AhmadGT
12/12/2022, 10:51 PMyes - g
Gosmokeless28
12/12/2022, 10:52 PMYes - t
tylerdurdenceketi
12/12/2022, 10:52 PMIt makes training much faster. You can stop the training if you want. Your graph is very good. - t
tylerdurdenceketi
12/12/2022, 10:53 PMWouldn't it overfit? - a
AhmadGT
12/12/2022, 10:54 PMits still sounds a bit all over the place - g
Gosmokeless28
12/12/2022, 10:54 PMNo, I think it would start to overfit if you train for 250 epochs and more. - a
AhmadGT
12/12/2022, 10:55 PMwell i reached 890 :) - t
tylerdurdenceketi
12/12/2022, 10:55 PMYeah wav file count matters I suppose. - g
Gosmokeless28
12/12/2022, 10:55 PMNot the wav file count, the amount of voice data - t
tylerdurdenceketi
12/12/2022, 10:57 PMCan you explain a little bit? Let's say I have 5000 wav files. How many epochs should i train? Is it too much to have 5000 files? - t
tylerdurdenceketi
12/12/2022, 10:58 PMMy last dataset has 7000 files - g
Gosmokeless28
12/12/2022, 11:01 PMThere's no such thing as too much voice data. But more to the point: You can have a dataset of 6 wavs that total up to 12 seconds of voice data. However, you can also have a dataset of 6 wavs that total up to 52 seconds of voice data. It doesn't matter how many .wav files your dataset contains, what matters is how much voice data there is in it. - t
tylerdurdenceketi
12/12/2022, 11:05 PMSometimes it warns that max decoder steps reached. Is it because the sentence is long or is there some other reason? - g
Gosmokeless28
12/12/2022, 11:05 PMAre you talking about the synthesis notebook? - t
tylerdurdenceketi
12/12/2022, 11:06 PMYes And inference in training notebook which is at the end of epoch. - g
Gosmokeless28
12/12/2022, 11:07 PMI don't know, but I don't think that's an issue. - t
tylerdurdenceketi
12/12/2022, 11:09 PMThanks. - g
Gosmokeless28
12/12/2022, 11:09 PMNo problemo - t
tangynacho
12/13/2022, 3:25 AMBy the way, I figured out my issue was that the frame rate of my samples was too high. I set them all to 20500 and the results are a lot better now - g
Gosmokeless28
12/13/2022, 3:31 AMOh, that issue - a
AhmadGT
12/13/2022, 5:19 PMwell i have 82 wavs they total up 6 minutes i trained the model for 840 epochs and its still sounds a bit janky - b
boiano
12/13/2022, 5:53 PMhow can i download tacotron 2? - a
AhmadGT
12/13/2022, 6:16 PMas far i as I know you can't download tacotron 2