Underdog demos 32MB speech model that also generates music — tiny on-device audio AI

Context

Underdog has shown off a speech model weighing just 32MB that can generate both speech and music, according to TechCrunch AI. Compact models of this kind could make voice and audio generation practical on lightweight devices and in low-bandwidth environments. Marketing teams experimenting with synthetic voiceovers, jingles, or in-app audio should note the capability, though the demo is early and claims are unverified.