What is Underdog's Woof voice model?
Underdog has released a small voice model that fits inside a browser game, a pointed bet that useful AI doesn't need a data center. The demo, built on a speech model called Woof, weighs only 32MB and runs locally on a user's device.
Why is Underdog building smaller AI models?
Most of the industry is chasing larger models, and those models demand huge amounts of processing power. Some researchers and startups are going the other way, building compact models meant to run on edge devices such as phones and laptops. Underdog is one of them.
The game puts players in a room with a retro tech theme. There is a typewriter, a CRT monitor and a desk lamp that clicks when you flip its switch. The opening screen warns players about “intrusive” cloud providers in the town, a jab at cloud-based AI models.
Can a voice model this small run without the cloud?
The interactions are simple, and that is the point. Users can type anything on the typewriter, and the game speaks it back. They can also rotate a music tuner to generate different kinds of sound. Everything runs in the browser through Underdog's speech generation model, which the company says can operate on most devices with no internet connection.
The model has 31 million parameters and is open for anyone to download. A model that small is unlikely to replace the large systems that power today's most capable chatbots. Still, it could be enough to drive local assistant features or generate personalized podcasts, which is the kind of use case Underdog is betting on.
The demo comes only weeks after Underdog launched Bark, a 167MB speech-to-text model also designed for edge devices. Bark handles transcription and dictation.
Sigil Wen, Underdog's founder, told TechCrunch that the work started with a personal problem. Instead of relying on notetaking apps such as Granola, he wanted to build his own system, one that uses AI models running on users' own devices. That way, people own their data.
“I think we will be able to run most capable models of today on the devices you already own within months,” Wen said. “We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI.” He was referring to his company's manifesto.
Underdog has also released a 27-billion-parameter model that can run on laptops. The company is additionally offering a waitlisted app that it is positioning as a locally hosted alternative to Muse and Instinct. Taken together, the releases show a company trying to cover the range from speech demos to a model large enough for a laptop, all designed to keep processing off remote servers.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.