A Glimpse of the Future
We're building FakeYou as just one component of a broad set of production and creative tooling.
About FakeYou
Your brain was already capable of imagining things spoken in other people's voices. This is a demonstration of how far computers have caught up. One day computers will be able to bring all of the rich and vivid imagery of your hopes and dreams to life. There's never been a better time throughout all history to be a creative than now.
Technology disclosure
We'll be happy to remove any of the voices featured here for any
reason. See our terms page for more
info.
”Deep fakes” are kind of like Photoshop when it first came out.
They're impressive, maybe a little bit scary, but they're about
to become the new norm. People will become accustomed to the
technology, and the results will be used mostly for creative
good, unlocking previously costly and unattainable high
production values for individual creators. It's our belief that
the next Hollywood will be you.
The technology to
clone voices is already out in the open, and the voices here are
built by a community of contributors. We're not the only website
doing this, and plenty of people are producing these same
results on their own at home, independent of our work. You can
see thousands of examples on YouTube and social media.
Even
if the United States chooses to ban this technology,
institutions in China, Japan, Canada, and other countries all
over the world are rapidly conducting and publishing research on
this topic. People everywhere have access to this easy-to-use
technology. State actors probably have even more sophisiticated
versions available.
As an interesting point, most if
not all of the voices produced by this website are actuallyLinda
Johnson with a little bit layered on top.
NVIDIA makes these tools publicly available for you to make
your own voice models.
Check it out!
FakeYou was previously Vocodes
FakeYou.com was previously known as https://vo.codes, but has been rebranded to better reflect its purpose and direction. (Not many people know what a vocoder is anyway, so it didn't make sense to keep the name.)
Thanks
Thanks to the following individuals (in no particular order) for
help with data gathering and annotation, Discord moderation, ML
advice, etc.
Vegito1089, Shin, Ashurath, MakaveliGH,
Blutarch Mann, Yahia, Tim Squid, Seuneramet, Matt, Seth,
CookiePPP,@r9y9.
The following papers, models, and
resources were used:Tacotron2 (BSD-3 license),glow-tts (MIT
license), MelGAN (BSD-3 license), Multi-band MelGAN
(paper),HiFi-GAN (paper),Wav2Lip (MIT license), CMUDict (BSD-2
license),LJSpeech (public domain),VCTK (CC BY 4.0).
Videos
are generated by the amazing Wav2Lip system (paper), by Prajwal
(@prajwalkr14), K R and Mukhopadhyay, Rudrabha and Namboodiri,
Vinay P. and Jawahar, C.V.
Contact
Reach out to "echelon" on Twitter and Hacker News. Say hi.
The Storyteller Company (registered as Learning Machines, LLC)
Built With
