Original Content

In the AI MEGA link, you can find instruction for using the google colab.
At this point, most of you know that Google has been doing some crackdowns on AI voice colabs, so I will give some instructions as to how to download a local program. Note that there is currently an effort to make the instlation and usage of RVC programs more user-firendly.

Firstly, you will need to download a RVC program. I will recomend Mangio, since it doesn't crash 99% of the time. Here is the Git page:
https://github.com/Mangio621/Mangio-RVC-Fork

If you are having toruble with the download, you can always type "RVC Mangio download" on youtube to find some help.

Once you are done with the download, run "go-web.bat" to open the web UI.

You will only need to know about three files:

audio - this is where you put the audio that you want to be edited by the AI.

weights - this is where you will put the the .pth file of your AI mvoice model.

logs- this is where you put your .index files. Note that here you will have to first creat a folder with the name of the voice model and then put the .index file inside that folder. For example:
Mangio-RVC-v23.7.0\logs\cocomodel\coco.index

As for ny other information as to how to use the UI, you will find links in the AI MEGA file folder "ai tools" or you will learn about it in the youtube tutorials.

Here is the AI mega link:
https://mega.nz/folder/wH1FxTqD#TaV9p40IymmrR9brpOq0KQ

I don't think I've ever shared these outside of /co/

I used that ElevenLabs free auto-translate for videos thing to make a bunch of lines of Vera speaking in almost every language. This is so that I could train a new model with some more data (it totals at ~21 minutes). I tried to clean and cut out the silence between words as best as I could, I'll probably put out a dataset where all the audio in this one is passed through Adobe Enhance so that it'll sound a little cleaner and less shit (Vera1/2/3NEW have already been passed through Enhance however). In either case I'm posting it here now because I still have to figure out how to train it since I don't know which program or collab to use now that Google killed off the decent collab links from 2 months back.
https://www.mediafire.com/file/fsgh7ixsyhq932c/verasetNEWER.zip/file

Apparently you can use this program called Replay to make AI covers, if any willing to try it go ahead because it not work on my laptop https://www.tryreplay.io/

>>1885
I will try it out.
I think that a zaryan already made a AI cover with replay.

>>1885
>>1886
I have installed it and wow. It's pretty easy to use! Here is a Coco vocal test using the AI model.

We should tell about it to other teens. I will make a tutorial on how to use it.

I was going to do a thumbnail, but I have little time for that.
The midi was not edited by me, but another zaryan. It uses the sound sources from here: https://docs.google.com/spreadsheets/d/1nj-a6mtLuWhhUCrbB9XktQE5DErR4RtMGtx0h4LPzE0/edit?usp=sharing
Also, I originaly intended to sing it myself. It turned out decent, but then a friend sent me a better singing record (maybe taken from Guitar Hero).

>>1881
hey guys i downloaded rvc-mangio 2 days ago and i started working on some ai covers
as you might have noticed by pressing the mp3 file the voices my pc generated are absolute trash could someone help me in making the voices sound less robotic and more "human" sounding??
i used the web-ui (go-web.bat) in the RVC-Mangio folder and used the voices with the most epochs
also my graphics card is a bit old and my computer crashed a few times if the mp3 sound file was too long

>>1889
You could try to thinker with the options. Try upping the voice tone higher (6-12). That usually fixes the lowpitch robotic sounds.
Some people have used some addobe programs or audacity to edit the voices.
If you still cant get good results, then the original voice could be the problem. Try singing yourself the parts that sound bad with AI.

>>1891
gem

>>1890
hey man i finally got the opportunity to work on smth again i listened to your advice and made this
if you have some tips on how to make the voice sound better even with another program that would be helpful

>>1893
(might have fucked up the voice sync)

>>1894
>>1893
It sounds pretty good!
Don't try to make it perfect, since AI has it's limitations.
I like it doe!

I don't think I have any other tips. Just experiment with editing and you will get great results.

>>1895
thanks man

caca

>>1897
'em

>>1900
I think that the pitch is too high. Tone it down a bit.
But I bloody love the idea!

>>1901
lower than that it sounds robotic

>>1901
actually wait i'll see what i can do

>>1900
>>1901

>>1904
I think it's better, even if it has some parts that sound a bit robotic.
Gem in the rough.

>>1904
>>1900
Most glistering gemerald all week, and likely for many more weeks to come

>>4171
how do you people make this
also mymy is shorter than maya and coco is the tallest (if i'm not wrong she is supposed to be like 1,8 meters tall)

>>1907
i think it was some sort of ai, not sure what was the name. check the ai zellig thread on /qa/, i think that you may find the name somewhere.

Testing out something and these two images have inspired me to make this (no I have not gotten to converting that middle part yet, sorry)

>>1909
Shining, glistening, zarty saving 'rald. I hope you finish it.

Maya can finally scream now

New Coco models trained with new data from ElevenLabs. Still in a bit of a testing phase, but this song used 180 epochs
DL: https://huggingface.co/filler5736/cocomodelNEWER/tree/main

>>1913
shining gem!!!
also question how did you get the voice to sound so clean??? i tried to use both RVC and Replay but the damn voice keeps sounding robotic
also could someone explain what difference has a model with low epochs and one with many epochs?

Bumo

>>1914
Might just be the instrumentals doing a good job covering up the shittier parts of the voice + mixing and reverb.
Epochs is just how long a model is trained to "sound" like the voice, the two I have available for download where ones I think were at a good stopping point before it went on to overtrain too much.

requesting maya sing this
https://www.youtube.com/watch?v=1aNPNcmp_2A

AIs are bad at hands and eyes just like real humans but at least they released the Coco model already https://pixai.art/model/1644300693524083429

Made a model of Deel 6 Vera when her VA (Kayleigh) dropped out after Deel 3 and Joyce (Mymy's VA) decided to take her role for Vera. I used ElevenLabs to generate a bunch of voicelines for data.
https://huggingface.co/filler5736/VeraJoyceVA/resolve/main/verajoycemodel.zip?download=true
I might do something fun with this, who knows.

Bvmp

>>1919

what song is that? its kinda of a bop

>>1921
summer breeze by piper
https://www.youtube.com/watch?v=mH5K6fm2C0I

>>1922
Thanks for bumping this thread. Now i can upload this gem:

Kino is back on the menu fellas
>Applio colab link so now you don't have to have a NASA computer to run it or wait 9000 years on huggingface:https://colab.research.google.com/github/iahispano/applio/blob/master/assets/Applio.ipynb

Someone a while back requested one of the girls sing Kali Yuga, so if that person is still here this is for you.

>>2276
Oh I remember this song, I wanted to cover it once but then forgot
<original.mp4

Another Vera AI cover has hit the towers

>>3454
Fact checked by multiple peer reviewed studies to be something she'd play and sing

>>3472
wonderful

(the AI shits itself if it continues past that part)

>>3454
Completely fucked up by not including the backing vocals by mistake, this should fix that.

>>3475
This should be uploaded in video format on jewtube

Silent Hill 4 - Your Rain

>>3501
This one worked out really well.

Tomorrow

NEET Maya

>>3576
Zaryan don't you know that truecelcore is literally fact checked to cause violence

>Zaryan don't you know that truecelcore is literally fact checked to cause violence

Falling Zellig

>>3579
Needs more tuning at the start

>>3580
not wrong, I simply could not figure out how to make it sound better, probably needed a better mymy model

>>3581
I guess you're using Applio, try these models
https://soyzellig.party/zellig/res/29468.html#29515
And if you can't really find a good solution, go to weights.gg set the Consonant protection to Safe and it should fix it

Moved to >>>/zellig/66912.