Underdog shows off a 32MB speech model in new demo

Wait 5 sec.

Despite its small size, Underdog's new model can generate both speech and music.