Underdog Releases 32MB On-Device Voice Model Woof
Understand the story
Startup Underdog released Woof, a 32MB voice generation model (31M parameters), on November 6, 2026, available as a free download. The company claims the model can run offline on most devices. The release comes with a retro-tech-themed browser demo game: users can type text on a typewriter to have the model read it aloud, or turn a tuner to generate different music. In the weeks prior, the company also launched Bark, a 167MB speech-to-text model. Founder Sigil Wen said the goal is to let users run models on their own devices and keep control of their data.
AI overview of reporting · Updated 1 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- TechCrunch · AIUnderdog shows off a 32MB speech model in new demo
Startup Underdog has released Woof, a 32MB speech generation model (31M parameters) that can be downloaded for free and, it claims, runs offline on most devices. The demo is presented as a retro-tech-themed browser game, where users can type text on a typewriter for the model to read aloud, or turn a tuner to generate different music. A few weeks earlier, the company also launched Bark, a 167MB speech-to-text model; founder Sigil Wen says the goal is to let users run models on their own devices and keep control of their data.
Story heat trend
Not enough continuous observations to show a trend yet.