As artificial intelligence (AI) continues to develop, the use of “deepfakes” on the internet is becoming more and more common. Although the use of these kinds of “deepfakes” to produce non-consensual porn videos and fake videos of celebrities and prominent personalities has become a huge issue over the years, there are still some AI-based deepfake generators existing on the internet. And this creepy deepfake voice generator is one of those tools that use AI to mimic the voices of various politicians and celebrities.
Developed by an Atlanta-based engineer, Brandon Thomas (@echelon), “Vocodes” is essentially a deepfake voice generator that uses an AI-algorithm to mimic the voices of famous personalities like Bill Gates, Mark Zuckerburg, Bryan Cranston and many others.
Now, according to the developer, “Vocodes” is not something serious but a simple tool to have some fun. Thomas built the tool as a side-project during the mandatory lockdowns and does not “mean to offend any parties”.
For creating the tool, Thomas used his and (probably) his partner, Linda Johnson’s voice with various layers on top. Then he utilized the open-source tools from NVIDIA to train the AI model of “Vocodes” to mimic the voices.
I Made Bryan Cranston Say “You Are Goddamn Right”
Now, I tried out the tool myself to see how it works and I must say that it’s pretty cool. The voices of the personalities are categorized based on their quality. There are four categories — Highest Quality, Decent Quality, Poor Quality, and Terrible Quality.
So, as a Breaking Bad fan, I had to make Bryan “Heisenberg” Cranston say “You are goddamn right” just for old times’ sake and I did.
However, in the first try, the model could not process the words. I had to try a couple more times to make it work and when it did, it was pretty impressive. Although the flow of the dialogue wasn’t there (obviously), the voice was pretty much accurate.
So, if you want to try out “Vocodes” for some fun voice-mimicking times, you can click/tap right here.
Awesome! Image-wise that’s just great!
But how about the voice logistics? What if you need to replicate the voice of an actor who’s not able to come to the studio? or if an actor with whom you started an animation project passed away? or if you want to use the voice of a famous character who’s really difficult (let alone expensive!) to be brought to the recording studio? Respeecher uses machine learning algorithms to do all of the above and more, which is a nice, indeed necessary addition to Cartoonizer’s video and image conversion.