648: VALL-E: Uncannily Realistic Voice Imitation from a 3-Second Clip

Text-to-speech gets a groundbreaking update with Microsoft’s VALL-E. On this Five-Minute Friday, Jon Krohn investigates how the Microsoft team modeled their tool to replicate natural human speech using just three seconds of a person’s voice. Additional materials: www.superdatascience.com/648 Interested in sponsoring a SuperDataScience Podcast episode? Visit JonKrohn.com/podcast for sponsorship information.