Underdog shows off a 32MB speech model in new demo

Underdog has demonstrated Woof, a compact speech generation model, through a browser game where users can hear typed text spoken aloud and generate music. The 32MB model is designed to run on devices without an internet connection, offering a possible route to private, locally run voice features as AI development often depends on substantial computing resources.

Woof has 31 million parameters and is available to download. It follows Underdog’s 167MB Bark speech-to-text model, released weeks earlier for transcription and dictation. The company says it has also released a 27-billion-parameter model for laptops and is developing a waitlisted app positioned as a locally hosted alternative to Muse and Instinct.