‘Project Gucciberg’ Creates AI Deepfakes of Gucci Mane Reading Classic Literature

Project Gucciberg

Photo Credit: MSCHF

Project Gucciberg is a deepfake AI developed to read classic literature in Gucci Mane’s voice.

It’s the latest development from MSCHF that uses machine learning to create the ‘audiobooks.’ The AI can read a selection of classic texts like Little Women and Beowulf in the trap rapper’s voice. They’re all free to listen to and come with book covers designed to look like Gucci Mane albums.

So why did MSCHF create Project Gucciberg?

Audio deepfakes are becoming especially common, and celebrity figures are always a target. The Tom Cruise deep fakes circling TikTok are just one example of the phenomenon. The only requirement to train an AI on speech patterns is lots of sample data. There are hundreds of hours of footage of people like Tom Cruise, Joe Rogan, and Gucci Mane.

Speaking to The Verge, MSCHF’s head Dan Greenberg detailed how Project Gucciberg got started.

Greenberg says MSCHF collected around six hours of audio of Gucci Mane talking. Sources included podcasts, interviews, and YouTube videos. The team then created transcriptions of the clips with the help of text-to-speech software. They created a Gucci Mane pronunciation key to better capture the idiosyncrasies of his voice.

“Gucci’s pronunciation follows a very particular cadence – he uses a much greater variety of vowel sounds, for instance, than your average TTS reader would,” Greenberg says. “The dictionary breaks words up into phonemes that our model then uses as building blocks.”

“For a simple example, we need our model to know what syllables to elide or flow into each other across words. It needs to know to say ‘talm ’bout for ‘talking about’ for example.’ Spelling out that phrase in the Gucci dictionary created for the project looks like { T AH1 L M B AW1 T}.

While the deepfake Gucci Mane is impressive, it’s not always cohesive. It sometimes sounds like he’s speaking through a bad quality mic, or over a low-quality stream. That’s because the training data may have featured those problems itelf. So why did the project creators choose Gucci Mane of all people? Greenberg says it came down to two reasons.

Gucci Mane’s distinctive voice and way of talking, and the Project Gucciberg pun brought the project to life.

MSCHF did not approach Gucci Mane, nor seek permission to use his voice in this way. It raises some questions about copyright, as noted by researchers. “We didn’t write the books, and we deep faked the voice.” Is this copyright infringement? Is it identity theft?