Underdog shows off a 32MB speech model in new demoTechCrunchNov 06, 2026 13:02ShareDespite its small size, Underdog's new model can generate both speech and music.