Underdog shows off a 32MB speech model in new demo Despite its small size, Underdog's new model can generate both speech and music. By Ivan Mehta ยท TechCrunch Despite its small size, Underdog's new model can generate both speech and music.