Underdog releases downloadable 32MB speech model with browser demo

Underdog has released a browser game demonstrating Woof, a downloadable speech generation model that occupies 32MB and has 31 million parameters. Users type into a virtual typewriter and hear the text spoken back; a tuner also generates different kinds of music. The company says the model can run on most devices without an internet connection. The release follows Bark, its 167 MB speech-to-text model for transcription and dictation on users' devices. Woof's browser demonstration shows local speech generation, while the broader claim of compatibility across most devices comes from the company.