Underdog demonstrates 31.7MB voice AI model 'Woop' in browser

Underdog released Woof, a 31.7MB voice synthesis model, and demonstrated it running entirely in a web browser, converting typed sentences into spoken audio without any installed software. The company said every voice in its browser-based demo game—set in a room with a typewriter, CRT monitor and lamp—was generated by Woof, and TechCrunch covered the demonstration. The release matters because it shows a compact speech model can run client-side on the web, sidestepping app installs and, potentially, server-side inference costs. What to watch next is whether Underdog publishes quality benchmarks against larger text-to-speech models and whether outside developers can access Woof beyond the demo.