15.ai

Read the changelog, he's been taken off because he's not quite ready

Yeah, he did build a Wheatley but chose to postpone it:
>Postponed due to instabilities during inference. A few Wheatley voices are scattered throughout this website.
You can hear him talk in Examples and the 404 page though

blob:fifteen.ai/8ce455d2-103a-4581-bae5-8223f3c2d9bc

voca.ro/ghHuH2dMPtz

vocaroo.com/3YQXRDyLhsy

?

Are you the same retard as
Because you are a retard, you are not as smart as you think you are

vocaroo.com/6FlNphwwuP2

Sadly no. You can find prefectly imitated voices of GLaDOS that are over 10 years old.
Someone claims to have written an AI that can identify and sound as much as these characters as possible with as little data as possible (which makes no logical sense btw).
The site is full of links to other "sciency" sites (trying to be convincing). It is very, very short in the description of how it could be done and there's some premature damage control going on as well.
Still hoping to hear some Cozy Glow though.
youtube.com/watch?v=RXOuzytkHGA

>top MIT researcher makes his AI usable for free online
>NEET on Yas Forums overanalyzes a single sentence and thinks they’re smarter than a fucking genius and can do better

ironic shitposting is still shitposting

voca.ro/4CaSe0LsZaA

>You can find prefectly imitated voices of GLaDOS that are over 10 years old.
Verifiably false.
> which makes no logical sense
Explain? Because it makes perfect sense. He’s trying to minimize the amount of data required to clone a voice. How is that hard to get?
> The site is full of links to other "sciency" sites (trying to be convincing
It’s a fucking bibliography and it links to actual ML papers
What the fuck are you actually on?

youtube.com/watch?v=widzHoMTu_U
vocaroo.com/nxRZVbVs9BY

youtube.com/watch?v=95BfgNHP58s

Attached: 066.jpg (680x598, 47.73K)

>It is very, very short in the description of how it could be done
Because he's putting together an actual scholarly paper on his process. Did you miss where he said he's working with MIT CSAIL?
>Someone claims to have written an AI that can identify and sound as much as these characters as possible
So what's your more logical explanation then, genius? He found voice actors for all these characters who are able to do perfect impressions and who are recording the requested lines live as they're submitted?
>with as little data as possible (which makes no logical sense btw)
Confirmed for not reading shit.
His approach uses dialogue from other sources to train the AI on how to speak English, while only needing a small amount of dialogue in order to train the AI on how to sound like the intended character. Existing approaches were focused on training the AI solely on audio from the person it would replicate, which would require upwards of 40 hours of it for good quality. When he said "very little available data" he meant available data of the intended voice.

Attached: 1534506522181.png (619x562, 112.33K)

just don't respond to bait, let the retard seethe on his own

Attached: rtfm.png (182x91, 3.65K)

voca.ro/55KDS29vB4A

I have a request. I need to go to sleep, and have been busy creating dirty talk lines, sound effects, and the like for use in a potential project. If an user, or user's are doing similarly, could you post a collection of your attempts at some point? I'll try and snag the link from an archive. Thanks to anyone who might respond in advance.

>Verifiably false.
It's not?
>Explain? Because it makes perfect sense. He’s trying to minimize the amount of data required to clone a voice. How is that hard to get?
Try to be like something as much as possible by knowing as little as possible about it makes no sense. This is an attempt to sound impressive and resourcefull.
>It’s a fucking bibliography and it links to actual ML papers
What the fuck are you actually on?
It's a spreadsheet with some links and nondescriptive text thrown in there. Ease down on the damage control the next time because it gives it away. We all know that it was just a prank bro. If there's a technical evolution of some sort it needn't be defended.

you are dumb neet

voca.ro/QgNNf5aY49S

vocaroo.com/jiqgufTbzJp

the irony of me being unable to make glados sound more robotic and inhumane is not lost i hope

Attached: .png (1280x720, 847.31K)

I transcribed all of this caption as best as I was able, but I don't have the patience to splice it all together and add sfx to it
dropbox.com/s/o2mldgzjaa3v3j6/glados_premature.7z?dl=0

Attached: 1583570115946.jpg (1700x1605, 1.1M)

vocaroo.com/jSEVeIyRIGi

Attached: cool.png (712x514, 794.35K)

I just don't like it when 4chins pokes fun of MIT. We all have some weak spots and holy grounds.

portal2sounds.com/

vocaroo.com/jpHa8ZPcIba

Lots of sexy potential with this one.

Voice transfer from minimal input data isn't a new thing. What exactly is your explanation of how this is being accomplished that's apparently more logical?

I've went through and created my own copy as well. It's different, though. I can edit it together sometime within the next day or two, if the threads are still up I'I'll post. Thank you.

vocaroo.com/6xioRCDvifA

NO GLADOS NO

voca.ro/ofOb0NjzHQh

Attached: 1400428407654.jpg (477x343, 52.67K)

voca.ro/8q6NSZdjfXk

Everyone follow 15 on Twitter and beg him to do TF2 merc voices
twitter.com/fifteenai

vocaroo.com/lNaRDQMHOyX

The only reason he did pony voices is because /mlp/ handled the workload of ripping, cleaning, trimming, and labeling all of the audio from the show. /mlp/'s creating their own AIs, but since the dataset already existed 15 cooperated with them and made use of it.

If you want TF2 voices, you and/or some other anons have to provide the voice lines and transcriptions for 15. There's info in the general on /mlp/ as to how to go about doing this and organizing it.

These are pretty much already compiled on the tf2 wiki, shouldnt be too hard\

God, I managed to get GLaDOS to say 'I love you' and my heart is melting.
Thank you for this, user. It's been something I've been dreaming of since I first saw her.

Attached: tumblr_2109447391c91fdc848ae2896cef22d5_ffb23b84_1280.png (754x1194, 168.54K)

>These are pretty much already compiled on the tf2 wiki, shouldnt be too hard\
That's great! I hope you're able to finish them soon.

don't thank me, thank 15, the MIT guy who made the whole thing and put it up on a website for us to enjoy

who knew that robot was so based.

And if you want some incentive to do that, imagine:
Gmod/SFM shitpost videos with complex custom sentences without the need for manual sentence splicing.

i'm not not doing it because its hard, but because i am very very lazy and dont really care

voca.ro/mGvqbSccayB

The amazing thing with our current AI is that it can try out a lot of things. If it covers a gazillions different versions of a voice and some guy listens to some of it and tunes it... that's not an actual intelligence at work there other than the programmer tuning it. It just covers a lot of ground.
Minimal input data is indeed not a new thing. On the contrary it's pretty much outdated. We needed it back when we didn't have so much space to move around with. A dying art form but I do enjoy it when my loading screens take less time so hopefully there is still some "room for it"(I hate my life).
youtube.com/watch?v=seFHSuL0nsw

Well, bless him then

Attached: tumblr_pran617Ic21uckub3o1_1280.png (1000x1002, 1.14M)

It'd be difficult to recreate SHODAN's voice with this tech, right? Given how glitchy and corrupted is her delivery.

voca.ro/5M2BRRyuTEs

voca.ro/ljPm6uetCdv

voca.ro/iBJsmL8fT8a

Attached: casserole.jpg (480x360, 9.89K)

voca.ro/HLug20IzAw3

>vocaroo.com/6xioRCDvifA

Attached: INeedADrink.jpg (560x564, 59.23K)

vocaroo.com/ob3MAMiVrcE

voca.ro/oeQ18NGJuSU

Attached: space.jpg (1191x670, 104.2K)

vocaroo.com/ilOMwJCnPFy

Attached: 1558130511181.jpg (1020x1200, 174.54K)

vocaroo.com/jRvxXaYZYQR
oops

What did she mean by this ?
voca.ro/h2kI1BYSGu0

Attached: wut.jpg (880x1280, 103.74K)

Attached: hersmileandoptimismgone.jpg (607x988, 91.04K)

old.vocaroo.com/i/s0S2hjqnqp78
I can't stop listening to it.

Attached: IMG_20180809_165625.png (554x403, 66.39K)

Why don't you go ask 15, the creator, yourself? Just from this post alone you can tell he knows exactly what he's talking about.

Making his what?

Shit, bad link.

robot waifu

robo waifu

>muh big words
I can google audio engineer jargon too.

voca.ro/8ApfZbJzG8F