Given that the guest was interviewed by podcaster Joe Rogan, who has been dead for more than ten years, it was unexpected. No, we are not inadvertently stating that zombies exist and that they predominantly affect former CEOs of large firms. Actually, both voices were produced by artificial intelligence (AI).
The podcast.ai podcast uses artificial intelligence to create all of its podcasts, and this week they used it to produce an unsettling (if not very comprehensible) dialogue between a bot trying to be Rogan and a bot pretending to be Jobs.
The team behind the bots writes on their website, “We wanted to push the boundaries of what is possible with current state-of-the-art voice synthesis, and we wanted to create content that can encourage others to do the same.” No one had a greater influence on the technological world than Steve Jobs, which is why we restored his voice in the first episode.
The researchers created transcripts of the dialogue using language models trained on Joe Rogan’s audio recordings and his biography, as well as recordings of Steve Jobs that were made available online. The text-to-voice generator Play. ht was then used to create astonishingly accurate voices from that transcript.
Rogan-bot introduces itself with “Hello crazy bitches” before forgetting the name of its podcast. Welcome to the Bro Jogan experience in yet another iteration.
Jobs-bot said, “It’s been a long time since I’ve been on the show; I’ve missed this, it’s always enjoyable [1:51],” before mentioning that he was no longer alive. The podcast had never included the actual Steve Jobs.
The end result is surprisingly convincing, especially in the text-to-speech department, barring a few audio glitches and instances where the algorithm veered off course (Joe Rogan, met with proof of the afterlife right in front of him, spends a surprising amount of time talking about LSD and Buddhism). However, if you listened to the transcript without knowing that it was generated by AI, you would think that both persons were conducting the interview while there was some kind of carbon monoxide leak because of a few little nonsensical flaws in the text.
The podcast allows listeners to suggest interviewers or interviewees for upcoming episodes and promises to produce more episodes with AI-generated text and voices.
