50% Error in rvc

[past relic]

anyone can help me with this??

just "trained" the model with Applio, and when i tried to use the new model with the new index i cant

[slender citrus]

G_233333.pth is a training weight

it is not a voice model

[past relic]

oh

how can i use it?

[slender citrus]

find the actual voice model

[past relic]

just using the index in the older VM?

[slender citrus]

example

red are voice models, D* and G* are training weights saved every 10 epochs

[past relic]

hmm

i dont have this

just this

[past relic]

this is in Applio

[slender citrus]

are you sure?

[past relic]

i think??

when the google colab is ended, i download the index in Applio

[slender citrus]

is to go logs/modelname

[past relic]

and where i can find this?

sorry, im new at this xd

[slender citrus]

that was the voice model

and you did not download it

check the backup folder on google drive

[past relic]

the second one?

[slender citrus]

yes

[past relic]

Ok, now i can

Ok, the training doesnt works fine XD

just the same voice with a baddest quality

[slender citrus]

based on the index size I say you fucked up something

show the logs for preprocess

[past relic]

how? xd

[slender citrus]

in colab

scroll up to preprocess outputs

[past relic]

i use this in applio

[past relic]

ok

[slender citrus]

not the UI

[past relic]

this?

[slender citrus]

more

[slender citrus]

I bet you did not provide the proper input to the preprocess

[past relic]

just this

[slender citrus]

you gave it a file name instead of the path

to the very top

[past relic]

hmm idk :c

[slender citrus]

did you use dataset creator?

[past relic]

Im trying to upgrade my model im using, bcs has a "english accent" and cant pronunce "R" (im spanish)

[past relic]

no

i think

idk

xd

[slender citrus]

how did you do the preprocess then

[past relic]

i just followed a YT tutorial xD

[slender citrus]

scroll up the training UI

[past relic]

this is the top

[slender citrus]

so... what do you expect from 10 seconds of audio?

[past relic]

The truth is that the tutorial made me split an 8-minute video of my voice speaking into 10-second clips, and only remove the parts where someone else is speaking or not speaking at all

[slender citrus]

if you want a better accent, find a spanish model in #1175430844685484042

nvm, for realtime you should not use an index

[past relic]

I tried it, but the quality of the voices with the modulator does not look good, I use them for GTA roleplay, and they are VERY artificial, or they simply do not fit my voice

[slender citrus]

you've followed a silly tutorial

[past relic]

https://tenor.com/view/i-don't-know-idk-idk-about-that-gif-7336250873949261946

[slender citrus]

8 minutes / 10s should be 48 files

you only had 1 10s file as a source

where's the rest?

[past relic]

just one?!

[past relic]

they was 17 :c

[slender citrus]

anyway.. take your 8 minute file, load it into audacity

cut the pieces of not your voice out

save as one file

load that file into preprocess

[past relic]

I don't know how to do that honestly xd

The truth is that what I did was a mix of chatgpt and the YT tutorial, maybe that's why I screwed up

[slender citrus]

audacity:

play the file, find the sections where it is not your voice, select the piece

press Del to delete

[past relic]

I don't have that program. Likewise, can I mix the clips I already have and that's it?

[slender citrus]

if those clips are 10s each, you only for 3 minutes of audio

[past relic]

yup

[slender citrus]

you not going to get anything good out of 3 minutes of audio

but you can merge them into one file and load that

or upload them all together

make sure you select automatic splitting

[past relic]

I have another 20 minute clip, mixing that one and the 3 I have, could I get something good? As I said, I just want to be able to pronounce the R correctly with the modulator, and to be understood when speaking, since it is strange.

[analog radish]

also, you're using an over year old version of original wokada

and prob using vb audio cable

[past relic]

you mean I should update the modulator?

[past relic]

yup

[analog radish]

that also causes crackling and random issues on windows, dont use yt tuts

[past relic]

I use it in GTA Roleplay (fivem) then I need the virtual cable

[analog radish]

yeah, there's a better one

-realtime

[gloomy geyser]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

• Original Wokada

Not suggested nor maintained, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[analog radish]

read up the first link, wokada deiteris fork, and it has vac lite too

[past relic]

this is not the same than vbaudio cable?

[slender citrus]

no

[past relic]

its better??

[analog radish]

yes

[analog radish]

no, it says specifically to not use vb audio cable

[past relic]

It's the same modulator, right?

this

[analog radish]

if with "modulator" you mean realtime voice changer, it's better

[analog radish]

you need to click the download nvidia on windows

[past relic]

yeah, sorry, im using google translate for this XD

this

[analog radish]

yes

[past relic]

ok

Thanks for the help guys <33
I wanted to improve my model but I really don't know anything about this beyond turning it on xD

[past relic]

what is "force_gpu_clocks"

??

[analog radish]

its optional, u should go in the folder and run mmvcserversio
else
check https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#gpu-idling for more info

[past relic]

This good??

advanced settings:

It sounds quite different, but I think it's a little better understood, I'm not using INDEX either

[analog radish]

extra: 2.7
uncheck sup1

[analog radish]

force fp32 mode: on
disable jit compilation: on
crossfade lenght: 0.15

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#reduce-more-delay optionally this for lower delay but its more complex

[analog radish]

@past relic any updates?

[past relic]

I've been trying it and it sounds much better, plus I have almost no delay.

But I have a question, is it paid? I read something about it being 30 hours a week or something like that, I don't know if I misunderstood

this

[analog radish]

Its not paid

[analog radish]

That's only the wokada deiteris fork kaggle

[past relic]

I mean I can always use it as many hours as I want, right?

[analog radish]

Kaggle is a cloud (remote good pc) service, it has limited time bc it runs on a remote PC, kinda like cloud gaming

[analog radish]

Yes, you're using it locally not on cloud

[past relic]

And is it normal that it no longer has so much delay? now I have less than half a second, before I had approximately 0.6 seconds

It really works much better, and when I read that I was wondering if I installed anything else haha

[analog radish]

i mean isnt it a good thing this program is better lol?

[past relic]

Yes, it's great haha, but I didn't expect so much improvement going from an old version to this one hahaha

Well, problems solved, thank you very much!

[analog radish]

well, is this solved now?

great then