UNDERDOG

Underdog Unveils a Browser-Ready 32MB Speech Model Running Completely Offline

Weighing in at just 32MB with 31 million parameters, Underdog's new voice model powers an interactive retro browser demo without cloud dependency.


While major AI developers chase massive cluster sizes and cloud-heavy architectures, a counter-movement of startups is prioritizing edge-native efficiency. Underdog has stepped into this space by launching a retro-themed interactive browser demo called Woof, powered by a speech generation model that is a mere 32MB in size.

Featuring a vintage aesthetic with a CRT monitor and typewriter, the demo allows users to type arbitrary text and hear it spoken back instantly inside the browser. Packing 31 million parameters, the model operates locally across most hardware configurations without requiring an active internet connection, serving as an explicit nod away from intrusive cloud providers.