After building this model, the team conducted a behavioral experiment to see whether the AI- or human-generated vocal imitations were perceived as better by human judges. Participants favored the AI model 25% of the time in general, & as much as 75% for an imitation of a

→ View original post on X — @mit_csail