[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/vt/ - Virtual Youtubers

View post   
  [?]In ghost mode, only threads with non-archived posts will be shown
File: 9 KB, 225x225, 下載 (17).jpg [View same] [iqdb] [saucenao] [google]
10779147 No.10779147 [Reply] [Original]

Has anybody ever thought of copying vtubers voice using voice cloning technology? I just downloaded a software called overdub and I prepared one hour video of gura speaking without background music. After all the preparation I found out that the software charges fee to use it. Does anybody want to give it a try?
If you succeed you can make Gura say anything you want.

https://m.youtube.com/watch?v=i49xQdI0upQ&t=14s

>> No.10779202

>>10779147
We await this video of yours.

>> No.10779659

>>10779147
Pay the fee hobo

>> No.10779782

>>10779147
Fucking hate software that deceives you into thinking you can at least try its main functionality for free for a bit.

>> No.10779881

>>10779147
If you attempt to do this, you ARE going to get Cover on your ass in no time flat if you upload it anywhere like YT btw. This is one of their "Do not cross" lines.

>> No.10779909

>>10779881
He should do it anyway. I want to see what happens.

>> No.10779930

>>10779147
>anon creates the final yab

>> No.10780027

>>10779881
Fuck off. We need to deepfake Gura. Imagine the meme potential

>> No.10780028

>>10779147
Thanks for letting us know how the final yab might appear, now it won't be as impactful retard

>> No.10780031

>>10779147
I did try a long time ago, not Gura obviously but the results wasn't that impressive.

>> No.10780055
File: 1003 KB, 404x347, 1625282334813.gif [View same] [iqdb] [saucenao] [google]
10780055

>>10779147
honestly /vt/ would use this tech to make their oshis say how much they love them rather than cause yabs.

>> No.10780061

>>10780031
Ah I think it was because I tried with japanese voices. Maybe using english would be better. Do it anon.

>> No.10780102

>>10780061
Better yet, why don't you make an easy google collab with the available code on github. Let other people just use it to upload voice clips and wrote down what they want their chuuba to say.

>> No.10780113

>>10779147
pay and let us see the results faggot

>> No.10780153
File: 17 KB, 512x344, death note 1.jpg [View same] [iqdb] [saucenao] [google]
10780153

>>10779147
>If you succeed you can make Gura say anything you want.
>"anon, i think youre very based"

>> No.10780263

You're gonna need a lot more than one hour of shitty quality voice clips to get anything more than a robotic slur.

>> No.10780314

>>10779147
wait
why the fuck isn't Vocaloid vtuber a thing yet???? What the fuck is Crypton doing??? why are they sleeping on this?

>> No.10780326

>>10780027
>Fuck off. We need to deepfake Gura. Imagine the meme potential

Why deepfake when it's easier to recreate the model + rigging?

>> No.10780404

>>10780314
because you can't create vocaloid music/speech live?
do you know how it works?

>> No.10780478

>>10780102
Welp, found the collab. You guys cab try.
https://colab.research.google.com/github/tugstugi/dl-colab-notebooks/blob/master/notebooks/RealTimeVoiceCloning.ipynb

https://youtu.be/1WN8Jhfd4uM

>> No.10780500

>>10780404
who the fuck said it should be live? You realise most major vtubing for a long time were self contained videos right?

>> No.10780617

>>10780500
It should be live. Why bother if it cannot achieve it's intended use.

>> No.10780681

You know, there is a certain dinosaur vtuber that sounds a LOT like Gura, Spiña Yaeba. I think she uses some sort of voice changer if I'm not wrong.

>> No.10780734

I posted this here because I got no money. Maybe some cool anons with money can try it.

>> No.10782857

No one wants to try it?

>> No.10783038

>>10782857
I want to try it what do I need to do

>> No.10783112

Imagine the day when Gura decides to move on from Hololive. Then technology is used to bring someone else to act as Gura with said voice synthesizer.

Fuck that shit. The real deal cannot be replicated no matter what anyone does.

>> No.10783126

>>10780681
You're right, they do kind of sound like Gura. More like Senzawa actually.

>> No.10783215

>>10779147
be aware that you can seriously get in trouble for this

>> No.10783221
File: 25 KB, 640x421, monk.jpg [View same] [iqdb] [saucenao] [google]
10783221

Imagine your life after cover sued your poor ass

>> No.10783237

>>10783038
Download overdub first, just Google it.
Get a video of hours of Gura's voice, and don't include background music.

>> No.10783451

>>10779147
sounds like a nice way to get banned or possibly even sued

>> No.10784365

>>10779881
As long as he is not profiting from it there is very little they can do apart from taking the video down.

>> No.10785302

>>10779881
all they can do is take down the video.

>> No.10785337

>>10783221
here's hoping they can sue someone from a third world country they've never heard of

>> No.10786944

>>10779147
Forsen's tech guy will have a perfect Gura tts by next month.

>> No.10787243

Anyone has the "Trash taste boys talking about fucking Mori/Ollie" deepfake voice thing?
Asking for a friend.

>> No.10790536

Don't just look at the negative side. Think about it, if you are a game developer and you want to make a hololive fan game but you don't have permission from en-managers or you don't have enough money to hire gura to do the voicedub. You can use this technology to dub your fan game.

>> No.10791816

Too much work

>> No.10792340

>>10791816
What work

>> No.10792863

>>10792340
Says it takes 24 hours

>> No.10792932

>>10779881
People use this shit to copy like Bart Simpsons voice and Mickey Mouse. Why do you think cover is more powerful than Disney?

>> No.10793041

>>10786944
This. The best deepfake ai is stuck on twitch with forsen and that is probably a good thing

>> No.10793055
File: 4 KB, 266x45, file.png [View same] [iqdb] [saucenao] [google]
10793055

>>10780681
>click on among us stream
>see pick related in chat

>> No.10793598

>>10793041
ZULUL OH

>> No.10793796

>>10793055
She still needs to vent out every once in a while

>> No.10793918

>>10793055
>Senzawa Archive

Anon....it's an archive fag...not HER

>> No.10794461

>>10779147
I want some gura blowjob asmr
I'll pay for it. Now do it anon

>> No.10795633

Let's say you have a voice changer that allows you to speak 100% like Gura, either screaming or talking or murmuring. Then you get a same live 2d model, then you create your own YouTube channel with the name as clone Gura. What will you say to your audience in your first video?

>> No.10795948

>>10795633
a

>> No.10795993

>>10779881
Couldn't this blow up in Cover's face?
If a voice is generated by a software they don't own any right to it even if it's similar.
You can find real people with similar voices to known celebrities who sometimes make a career with this.
It would be the same as trying to patent a regional language because a vtuber used it first.

>> No.10796075

>>10779909
Video taken down and OP gets a lawsuit.
What are you expecting?
Maybe gura says not to believe any videos are real unless they are hosted on a cover channel.

>> No.10796131 [DELETED] 
File: 76 KB, 779x1091, 1620741426919.png [View same] [iqdb] [saucenao] [google]
10796131

>>Senzawa Archive
>Anon....it's an archive fag...not HER

>> No.10796274

>>10779147
>Has anybody ever thought of copying vtubers voice using voice cloning technology?
Everyone in Cover and whatever the fuck AI Channel's company is called, for starters.
At some point at least one chuuba company is going to go under for trying to pull this and getting sued out of existence.
My money is on PRISM Project since they're pretty dodgy (CEO is openly and loudly into NFTs for example and small enough they wouldn't be able to just eat a lawsuit.

>> No.10796468

>>10795993
>If a voice is generated by a software they don't own any right to it even if it's similar.
Nice try but Vocaloid is 17 years old at this point. There are no new legal loopholes being discovered here.
That being said.
>>10792932
>>10784365
>>10785302
This.

>> No.10797501

>>10779147
Roa

>> No.10800040

That sounds interesting.

>> No.10801882

>>10800040

Yeah.

>> No.10803489

>>10779881
"Examples of fair use in United States copyright law include commentary, search engines, criticism, parody, news reporting, research, and scholarship. "
Just make it a parody they cant touch you. Godspeed anon.

>> No.10808988

>>10780681
Chuuba impressionists are the future. for the fraction of an akasupa (You) too can be Gura's little Pogchamp.

>> No.10809145

>>10780055
suicide rate drops to 0

>> No.10812541

All this talk of doing it and how it's fine yet I'm not seeing any results here

>> No.10812638

>>10779147
Make her say “I love big nigger dick in my cunny.”

>> No.10812847

>>10803489
The parody has to be an actual parody though so something that Gura would actually say. It becomes a gray line if the parody causes tortious interference. Might claim it to be free speech but if you make it say something like "Hitler did nothing wrong" and push it as a real thing and it causes her brand to be damaged they could probably successfully sue for tortious interference since no court is going to reasonably assume pushing rrat that Gura is a secret nazi is a parody of anything.

>> No.10815967

>>10812541
Because this is a bait thread.

>> No.10822078

>>10779147
>Has anybody ever thought of copying vtubers voice using voice cloning technology?
No

>> No.10822280

>>10779147
Hololive fans are literally bronies.

>> No.10822714

>>10793918
I thought It was someone who dedicates his channel to archive Senzawa content pointing out how said vtuber was kinda sussy.

>> No.10822838

>>10780681
ah damn it, she already quitted a week ago? fuck.

>> No.10824051

>>10780681
Yo, she even uses a Gura skin for L4D, too weird.

>> No.10824549

>>10812847
>so something that Gura would actually say
Nope stfu

>> No.10824885

>>10812541
People can barely properly remove the background noise or music from their voices as it is.

>> No.10826953

>>10803489
A person's voice isn't copyrighted. Thus fair use doesn't protect you against impersonation and defamation claims.

>> No.10827170

>>10779147
I thought faking Gura's voice to make her sing a reflect Minecraft song parody would be funny but I have no idea how to even start doing something like that

>> No.10827274

Ken, this reminds me of the AI generated Joe Rogan chimp ice hockey team video

>> No.10827303

>>10779147
if you're raped i prison i'd vote to acquit

>> No.10827736

>>10826953
You're right we should all sue the kid that did impressions of us in middle school!
Or alrc baldwin from saturday nighr live for doing Drumpf!!!

>> No.10829468

>>10780314
Some people consider vocaloid VTubers, I don't know.

>> No.10830133

>>10827170
I already said in the first post. Get an overdub, then get an hour of Gura speaking video without background music. I was here looking for help because I got no money. After the preparation you let the machine learn Gura's voice then you can type the text and the machine will read the text with Gura's voice

>> No.10836333

>>10824051
Can you give me the timeline of the video where is she used Gura's profile?

>> No.10836365

>>10824051

Skin*.

>> No.10841216

>>10827274
post it

>> No.10842432

>>10783221
If you were already poor it wouldn't change too much

>> No.10842884

>>10841216
This one?
https://m.youtube.com/watch?v=8rCHad-N0o4

>> No.10843049
File: 390 KB, 1005x1020, cover.png [View same] [iqdb] [saucenao] [google]
10843049

>>10779147
>Gura
Why do that when you can get a Japanese vtubers voice to speak to you in English?

>> No.10843863

>>10843049
What did you mean?

>> No.10845145

>>10842884
Amazing.

>> No.10845220

>>10779881
Imagine being this much of a boot licking faggot

>> No.10845465

>>10842884

If we voice clone Gura's voice and then we generate this quality. I wonder how Gura will react once she see the video.

>> No.10846252
File: 65 KB, 505x475, 1605416615614.jpg [View same] [iqdb] [saucenao] [google]
10846252

>>10842884
Thanks for pulling that up, Jamie! I miss Roganposting...

>> No.10847150

>>10796075
>Maybe gura says not to believe any videos are real unless they are hosted on a cover channel.
That is big, that would kill clippers.

>> No.10848602

So, has anybody tried it yet?

>> No.10848722

>>10848602
I tried the demo on that google collab, but it didn't come out well.
https://files.catbox.moe/7a8hhc.wav
I probably need to set it up on my PC and train it properly.

>> No.10848909

>>10848722
Checked and keked, definitely not Gura but it's a start

>> No.10849872

>>10848722

You need hours of Gura voice for the training.

>> No.10849936

>>10848722
hi kiki

>> No.10850126

>>10779881
how does the boot taste anon?

>> No.10850249

>>10848909
I don't think I'm the guy to make this happen, I'm too busy right now.

>>10849872
And I'm pretty sure it needs to be without background music. Can't just feed it a stream.
It's not hard to find samples of Gura without BGM, it's just more work.

>> No.10850437

>>10780027
That's just begging for disaster.

>> No.10850493

>>10850249
Theoretically, if you have hours of gura with the same BGM throughout, you could clean out the BGM by just subtracting the BGM waveform.

>> No.10851050

Now let's wait for some anons with free time.

>> No.10851423

>>10843049
Imagine one of the subbed clips, but with the voice reading the subtitles.

>> No.10851506

>>10779147
It's extremely easy actually, look at forsen's tts, you just need a neural network trained on their speech with text captions

>> No.10851704

>>10851423
Replicating the tone would be another thing though.

>> No.10851856

I want to put a button activated sound box in my plush Gura

https://files.catbox.moe/4a3q3r.jpg

>> No.10852193

>>10851423
Would you give me some examples of the subbed clips?

>> No.10852301

>>10851506
Since it is so easy for you, why not make some cool video showing anything you want Gura to say?

>> No.10855603

>>10779147
Powerful idea

>> No.10856324
File: 77 KB, 536x400, hate.jpg [View same] [iqdb] [saucenao] [google]
10856324

>paying for software instead of harnessing 4chan autism to train and deploy your own ML model
you guys are officially worse than /mlp/

>> No.10856466

>>10850126
>>10845220
This is fucking ironic because if someone used technology like this to create fucking fake news and have lefty politicians say shit that could lose them voters you fucks would immediately call for regulations to jail people doing this.

>> No.10856581

>>10856324
Not even a meme in this case, considering that /mlp/ has quite literally created entire machine learning training datasets and models that do exactly this at a significantly higher quality than whatever ripoff some asshat on Youtube is trying to get you to pay for.

>> No.10858412
File: 155 KB, 512x512, 1633391068981.png [View same] [iqdb] [saucenao] [google]
10858412

Is it even possible to do unsupervised learning to "clone voice data" of a person? I thought at this point in time most speech synthesis models required a dataset of captions. Where are people going to get that?

>> No.10867397
File: 503 KB, 640x692, korone :L.png [View same] [iqdb] [saucenao] [google]
10867397

You dummies, all you guys need to do is label Gura's voice and send it off to the guy who does these.

https://15.ai/contribute
>tl;dr: If you send me a dataset of your favorite character/waifu/celebrity/whatever, I might include them in the next batch. Due to technical reasons (approximately uniformly distributed vocal frequencies), high-pitched/feminine voices work best.

Work would need to be done to isolate her voice (i.e. only labelling supa streams with known BGMs so you can remove the background music) and to streamline the labelling process (maybe a webapp that lets you transcribe in sync with VODs).

>> No.10872161

>>10780314
While we still don't have a proper Vocaloid Vtuber, we had some Vocaloid/Voiceroid Let's Plays in the past.

https://www.youtube.com/watch?v=1d1ygvUutz8&list=PLtVPi_p4lZ306FaqkBVLVhBGw2rqmLiDa

>> No.10873041

>charges you after you use it

I hate this shit, basically pretending it's free and wasting your time as you get something done and then it's like "you want the thing you made? Pay me bitch!".

>> No.10875241

>>10779659
No

>> No.10879511

Any more resources for voice clone? Better if it is free.

>> No.10883796

>>10867397
15.ai won't work. Recreating Gura's voice is unethical, even if he's tricked into doing it he'll just take it down when he gets Cover's letter.

>> No.10884490

Watch this video, if we replace the SpongeBob voice with Gura's voice it'll be so funny and cool right?

https://m.youtube.com/watch?v=hRlo8_13Zyw&t=39s

>> No.10885021

It is not necessarily a bad thing, look at the derivative work by dudul.
https://m.youtube.com/watch?v=XpbNxG1cd3Y

If we get the Gura voice machine we can watch more creative videos like this one.

>> No.10885968

>>10885021
Or we can get voice lines such as "Hitler did nothing wrong" and "Fuck my cunny daddy".
It's not just about Gura, if it can be done for any vtuber it will create unease.

>> No.10886942

You fuckers act like this is some novel shit, Hatsune Miku, voiceroid and heck even cheap TTS have been out there for ages.

>> No.10887016

>>10886942
none of those sounds like gura

>> No.10887172

>>10886942
Yes, but now there's the possibility of it sounding realistic

>>
Reply to thread [?]Don't expect anything heroic. Your post will not be uploaded to the original board.
Name
E-mail
Subject
Comment
Action