But you can hear the Schwarznegger.

and of course the classic line from the terminator 'I'll be back!'

Hi, I just found this group from Reddit and I’ve been curious to how people do voices, I’ve got a software dev background and a decent rtx pc, what do I need to do to make my own voices or improve some of the ones on the site?

Also I noticed some of the voice models are up for download here, are all the voice models available by request?

Pretty easy to make a voice,

Easy but time wasting.

You don't even need an actual GPU, we do it on Google Colab

What’s the notebook link?

First.

You need to prepare a dataset.

check #841422801965416538 channel for more

Will increasing the amount of audio examples make the model result less noisy or will increasing the training time do that?

both, but if you don't have enough audio then increasing the training time will just make it hyperfocus on the few examples it gave you

hmm it takes 2 minutes to train an epoch but the total time is 3 minutes wonder how much of that 1 minute is validation

Don't think I should distribute it before asking for permission

Wow what for me it took like 10-15 seconds per epoch

it depends on dataset size

Is that because I had a bajillion audio files?

Ahhh

i have ~300 files

68mb total

so like 25.5 minutes

though some of that is silence that gets trimmed

point is it's a different amount

and an epoch is a full run through the dataset

I saw the word overtrained once in the tutorial

Can someone define that?

more generally known as overfitting, it's where you train the network for so long that it starts to memorize the specific training samples you gave it, at the cost of being able to speak anything else

huh, and to not do that, how long do i train for