Underdog Demos 32MB Speech Model With Browser Game

AI companies around the world are competing to build larger and more powerful models, but those systems demand enormous processing power. Some researchers and startups are taking the opposite approach, building small models that run on edge devices such as phones and computers.

One of them, Underdog, has released a game built with its new voice model, Woof, which is only 32MB in size. The game features a room with a retro tech theme, a typewriter, and a CRT monitor, plus a desk lamp that makes a sound when its switch is clicked. Its intro warns users about "intrusive" cloud providers in the town, seemingly a jab at cloud-based models.

Players can interact with the typewriter and a music tuner. Anything typed on the typewriter using the keyboard is spoken back by the game, and rotating the tuner generates different kinds of music. All of it works in the browser through Underdog's speech generation model, which the company says can run on most devices without an internet connection. The model has 31 million parameters and is open for anyone to download.

While such a small model likely won't replace bigger ones, it may be enough to power local assistant experiences or personalized podcasts. The demo arrives weeks after Underdog launched Bark, a 167MB speech-to-text model also designed to run on edge devices for transcription and dictation.

Sigil Wen, the founder of Underdog, told TechCrunch he started building the app because, rather than using notetakers like Granola, he wanted his own system built on AI models that run on users' devices and let them own their data. He said he thinks most capable models of today will be runnable on devices people already own within months, and that the goal is to train the best open, American-made models that are capable and allow people to have private AI.

Besides the two voice models, Underdog has released a 27-billion-parameter model that can run on laptops, along with a waitlisted app the company is positioning as a locally hosted alternative to Muse and Instinct.

Source: TechCrunch