Elevenlabs.io - AI Generate anyone's voice with a minute of audio saying how much they hate trannies

  • 🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
it
Meowstery said:
This is so funny, I love making stupid memes with this AI voice but it's going to get stomped out so soon you can already see people panicking.

As an example of how incredible it is at replicating voices, here's a voice sample someone made on the /g/ board.
View attachment 4368500 And here's what the AI churned out using only that audio, that 20 seconds of talking.
View attachment 4368508


Incredible and scary, but what can it do with a lot, a lot more samples?
View attachment 4368520

Anyway, take this one, it made me giggle.
View attachment 4368544
This AI is going to be a treasure trove for coomers looking to make waifu jerk off audio clips, but it can't do too much with odd speaking styles like mario or certain cows like chris-chan
All these clips are made via AI if you couldn't guess
Click to expand...
can't do anything. you need A LOT OF samples or otherwise ai will be confused a lot. and its manual.
 
maybe right now my cpu is shit. but just you wait...

1707204650232.png
 

Attachments

Book 5 done, 3 more to go:

Dr. Matthew C. Harris is a coomer and would like you to know all about it. Also Zelda is his waifu.
 
  • Winner
  • Semper Fidelis
Reactions: Vecr and NSJAP
Book 6 done, 3 more to go.

For some reason I thought there were 8 books total, while there is actually 9 of them.

Also

based nigger
 
  • Winner
  • Semper Fidelis
Reactions: NSJAP and Vecr
Sprate Header said:
This was posted on the sharty earlier today, and hoooooly shit this is bad.
View attachment 4368525

I would be surprised (but unimaginably thankful) if fake "discrete recordings" of public figures saying insanely unhinged shit didn't start making the rounds during this next election cycle. And keep in mind, this Eleven Labs thing is what's publicly out there - Lord only knows what's being (or already been) developed in private.
Click to expand...
This is already happening with Sassy Justice and a twitch channel that shows very convincing deepfake debates
 
Book 9 done:

And that's it. Gotta say, having another video card really helped speed things up - while I was editing one chapter, I had a separate machine generating the others. This is why it only took me 4 months to complete this, as opposed to initially expected 9 months.

Also, here's the entire thing, all 39-and-a-half hours of it (uploaded to google drive, because it is impossible to upload 850 MB of audio here):
 
  • Semper Fidelis
  • Like
  • Political Sperging
Reactions: NSJAP, Catgirl Fornicator, RadioactiveMonkeyMan and 2 others
IamnottheNSA said:
Book 9 done:
View attachment 5948137
And that's it. Gotta say, having another video card really helped speed things up - while I was editing one chapter, I had a separate machine generating the others. This is why it only took me 4 months to complete this, as opposed to initially expected 9 months.

Also, here's the entire thing, all 39-and-a-half hours of it (uploaded to google drive, because it is impossible to upload 850 MB of audio here):
Click to expand...
I would pay money for this glownigger to get punched in the throat.
 
  • Autistic
Reactions: BirdUp
IamnottheNSA said:
Book 9 done:
View attachment 5948137
And that's it. Gotta say, having another video card really helped speed things up - while I was editing one chapter, I had a separate machine generating the others. This is why it only took me 4 months to complete this, as opposed to initially expected 9 months.

Also, here's the entire thing, all 39-and-a-half hours of it (uploaded to google drive, because it is impossible to upload 850 MB of audio here):
Click to expand...
Can you share us with your workflow? What software you used and how.
 
  • Like
Reactions: Vecr and Cowboy Kim
NSJAP said:
Can you share us with your workflow? What software you used and how.
Click to expand...
Well, first thing that I did was download the manifesto and use ABBYY FineReader to convert it to .docx, so that I could more easily grab pieces of text and edit them when necessary - while I decided to keep the typos in, I found out pretty soon that punctuation can influence the output pretty significantly, so I had to edit that in some places to make it sound more natural.
Then I installed TorToiSe TTS (or, rather, I installed this package, which included a webui) and started training a Dec voice model. The training set consisted of a mix of Dec's TO ALL JUDGES rant and 11.ai generated output of Ethan Ralph's "Smoke" blogpost, all broken up into a hundred of short soundbytes.

And so, once the model was done (training it took about a day), I got to generating. Set both Samples and Iterations to 486, temperature at 0.05. Only a single candidate, to save time, and even then generation of a single second of audio took a whole minute of real time - with Deepspeed enabled! I had it generating output overnight, and I was editing it throughout the day (with the secondary server generating pretty much 24/7).

Also, as I said before, once the output was done, I had to edit it (and often re-generate parts of it). I used Adobe Audition CS6 for editing - I've been using it for over a decade now and I'm very familiar with its functions.

Also I made a preset that makes the audio sound like it came from an old tape, just like the Dec tapes. I felt that the output sounded much too clean to be authentic, so I artificially-aged it like that (which also helped compless it so that I could upload it here). For example, here's Dec's ONLY HOPE FOR A FUTURE rant (the version everyone is familiar with, despite the higher quality recording being available):

And here's my artificially-aged version of a high-quality recording of said rant:

Pretty close imo.

As for the second video card - I pulled out an old computer that I had, inserted the video card there, installed headless Linux onto it, connected it to my home LAN and used it as a secondary AI generating server. Connected to it via Telnet, got files to and from it via FTP.
 
Last edited:
  • Like
  • Informative
  • Winner
Reactions: Relinquish, Windblown, Cowboy Kim and 2 others
It seems they recently censored the AI dubbing part of the website. Before this you'd get some hilarious translations, with some ENG to RUS ones becoming memes of their own, like Chester Stone vids for example:


...but now, whenever it detects a naughty word like "nigger" it doesn't even try and substitute it with something similarly vulgar yet inoffensive, but literally says "censored".
It truly feels like an unwritten rule at this point, that any and all new public web AI solutions will progressively get more and more neutered the longer they're up and running.
 
  • Informative
Reactions: NSJAP, IamnottheNSA, Niggolash and 2 others
You must log in or register to reply here.
Back