oh it sounds like crap haha

not english

do you havea screen recorder

?

I can upload a file

ok

do it

wich Graphics card do you have

?

I train usually on an NVIDIA V100

i had a nvidia gtx 560 ti

but its buggy graphics

less powerfull

but still works

lol

yeah haha

did you type numbers?

nope haha

he just only learned to say numbers

he act like 1 years old

yeah, I'm working on a speech embedding model so I'm hoping that that will allow synthesizing voices with much less data

in this case I think there probably isn't enough diversity in the dataset

wich speech are you working on?

I'm trying to reproduce this paper: https://arxiv.org/pdf/1910.10838.pdf

it is similar to Tacotron2, but it incorporates speaker embeddings which are independently trained on a large, unlabeled multispeaker dataset.

this is only demo

i use this most but sounds bad

how'd the michael stevens model sound when you did that?