A tiny voice AI that fits in a browser game can talk and make music

A startup called Underdog has released a voice AI model named Woof that is only 31.7 MB, small enough to run inside a web browser. It can turn typed text into speech and generate music. The company says it works on most devices without an internet connection.

What the demo does

Underdog showed it off through a small game set in a retro room with a typewriter, a CRT monitor and a desk lamp. Type on the typewriter with your keyboard and the game speaks your words back. Turn the tuner and it makes different kinds of music. The intro pokes fun at "intrusive" cloud providers, a jab at AI that runs on remote servers.

Woof has 31 million parameters, a rough measure of a model's size, and anyone can download it. A few weeks ago Underdog also released Bark, a 167 MB model that turns speech into text. It has a 27-billion-parameter model that can run on laptops too.

Founder Sigil Wen says the goal is open, American-made models that give people private AI. He expects most of today's capable models to run on devices people already own within months.

What is not known yet

The material does not say how natural the voices sound or how Woof compares with larger models. TechCrunch notes that such a small model likely won't replace bigger ones, but it may be enough for local assistants or personalized podcasts. It is also unclear which devices count as "most."