Underdog releases 32MB offline speech model Woof

Startup Underdog has released a new demonstration of Woof, a 32MB speech generation model with 31 million parameters, designed to run locally on phones, laptops and other edge devices. The demo, built as a browser-based retro-themed game, lets users type text that is spoken aloud and rotate a music tuner to produce sounds, all without an internet connection. The model is open for download and signals the company’s effort to bring capable AI to everyday devices, even if it is unlikely to replace larger cloud-based systems. Underdog founder Sigil Wen said the project is intended to give users private, on-device AI and control over their data, rather than relying on cloud providers. The release follows Underdog’s launch of Bark, a 167MB speech-to-text model for local transcription and dictation, and complements a 27-billion-parameter model the company says can run on laptops. Underdog also has a waitlisted application it is positioning as a locally hosted alternative to Muse and Instinct.