A.I. model trained to rap in Canadian lecturer’s voice
A new “deep fake” edit remixes the voice of clinical psychologist Jordan Peterson and dubs it over the popular Eminem song, “Lose Yourself.”
The result is a convincing yet deeply disturbing audio depiction of what Peterson might sound like performing the hit at karaoke.
YouTuber Coding Elite says by integrating two different speech synthesis methods, he was able to train an AI model to perform the song in the Canadian clinical psychologist’s distinctive voice.
“This model was trained using an implementation of the papers ‘Style Tokens: Unsupervised Style Modeling, Control and Transfer in End-to-End Speech Synthesis’ (https://arxiv.org/abs/1803.09017) and ‘Towards End-to-End Prosody Transfer for Expressive Speech Synthesis with Tacotron’ (https://arxiv.org/abs/1803.09047), using about 6 hours of clear Jordan Peterson audio,” Coding Elite writes in his video description. “Given a reference audio, this model will synthesise speech in that particular style.”
The YouTuber says the Peterson model is his best yet, and has superior audio compared to his previous attempts, which similarly overlaid the voices of Presidents Barack Obama and Donald Trump on “Lose Yourself.”
Recently a disturbing mash-up of actor Steve Buscemi’s face superimposed over actress Jennifer Lawrence’s Golden Globe acceptance speech showcased the ultra-realistic and unsettling nature of deep fake edits.
I've gone down a black hole of the latest DeepFakes and this mashup of Steve Buscemi and Jennifer Lawrence is a sight to behold pic.twitter.com/sWnU8SmAcz— Mikael Thalen (@MikaelThalen) January 29, 2019
Previous deep fake videos showed how advanced AI models can convincingly put words in the mouths of powerful world leaders.
