sorry I haven't gotten to your voice yet.

No problem

I have a round of general purpose improvements I'm working on, and then I'll work on adding it (and probably a couple other voices).

Add Voice Suggestion Chanel on your discorf server

Good idea. just added #794028287428132884

Good

Does uberduck.ai use tacotron?

Or else

it does use tacotron

if you train your own model using nvidia's tacotron repo, we can host it. that's what we're currently doing with the yandere dev model on there now (it was trained by a couple other people from the vocodes discord)

I have amd

you can do it on google colab if you want. it can be quite a time commitment though - no worries if you don't want to do it

This discord has some great resources on how to do it if you're interested https://discord.gg/Er4Sjq6

My gpu is high end

Saw your post in the audio deepfakes server about AMD - you can train on Google Colab with an NVIDIA GPU for free (or a better GPU for $10 / month).

There is a voice cloner on colab and it sounds doesnt like the orginal character

It can be kind of tricky to get things right. Feel free to post samples here and I can try to help.

Michael steves is actually vsauce did you know?

I tried training the michael stevens model ... unfortunately it didn't work. I will give it another shot once I've finished the work to train on smaller datasets. I'm pretty sure the actual size of this dataset is fine, but the issue is the lack of variety in the language.

Why didnt work did the audio sound not good?

Do you use tacotron waveglow or any else?

tacotron2. the model way overfit the dataset.

it reproduced tone and prosody pretty well but he couldn't really speak english apart from numbers

I found a colab of tacotron2

How does sounds like

I don't do this training stuff, but do you start with a pretrained model when training the stevens model?

yep, you need a pretty large dataset to train from scratch.

hi

how does the michael sounds like

bad quality