Voice AI startup Underdog showcases 32 MB model Woof

Voice AI startup Underdog has unveiled Woof, a text-to-speech and music-generation model roughly 32 MB in size with 31 million parameters, released as a free download and demoed through a browser game. Users can type text to hear it spoken or adjust knobs to produce different styles of music, with Underdog saying the model runs offline on most devices. The launch matters because it shows capable speech and music generation can be packed into a tiny footprint that runs locally rather than in the cloud, potentially lowering costs and easing privacy concerns for voice AI deployments. What to watch next is whether developers adopt Woof for on-device apps and how its output quality compares with larger hosted models.