Experiments · Prototypes, 2024
Podcast Machine & Digital Double
From a topic to a talking digital human: an automated podcast pipeline into NVIDIA Omniverse and Unreal Engine.
Type in a topic, press a button, and a digital human starts presenting a podcast about it.
The Dreamlab Podcast Machine was a prototype from early 2024 that chained the whole thing together. A large language model writes the podcast script, ElevenLabs gives it a voice, and the audio goes into NVIDIA Omniverse Audio2Face, which generates the lip sync and facial animation in real time. That drives a digital character in Unreal Engine, so the topic comes out the other end as a performance rather than a text file.
A quick, scrappy build, filmed on my phone off the monitors, so forgive the wobble. It was a useful way to find the joins between the tools, and an early look at how automated content and digital doubles might fit together.