means your audio is either in 44khz or stereo

several ways to fix that, but assuming you're not a nerd i recommend the following: - open audacity - select all the files and drag them in - if they're stereo, press ctrl+a and from the top menu do

tracks

→

mix

→

mix stereo down to mono

- at the bottom set

project rate

to 22050 -

file

→

export

→ `export multiple`; format

wav [...] 16-bit pcm

, name files

using label/track name

, select a new folder so you don't overwrite the originals just in case,

export

truly the speech

ah yes

Hello. Could someone explain to me what it is and what the Hifi-gan fine-tuner is for?

hifigan is the vocoder that tacotron2 and talknet use

finetuning can get rid of a robotic or flat sound given by universal hifigans

Hi guys! was trying to fine-tune TransformerTTS https://github.com/as-ideas/TransformerTTS model using the LibreSpeech dataset, can't figure out how to go about it though, any help on this would be appreciated.

embrace the compression

why use deepfakes when you could use an ai motion capture and the actual model

Lemme know if you can do that

Would love to do a comparison

i cant because i suck at figuring stuff out but i can send the ai motion capture thing

Ah

No need

ok

theres multiple

i saw one be used awhile ago but it sucked but theres probably way better ones

The one above just uses a source image

yeah

i personally think ones that turn footage into a 3d animation that you can apply to any model is better

There are many other approaches

because you can fine tune it and swap thye model easily

and use non human models

and it can be ported to games

(theres alot of reasons)

you can add lighting since its 3d

its just better to me