Underdog demonstrates a 32MB model for speech and music

Underdog has presented an interactive game built around Woof, a speech model measuring 31.7 MB. The startup says the system can generate speech and music, has 31 million parameters, and can run on most devices without an internet connection. The model is available for download.

The browser game depicts a retro-styled room with a typewriter, CRT monitor, desk lamp and music tuner. Users can type text for Woof to read aloud, while rotating the tuner produces different kinds of music. The introduction warns about “intrusive” cloud providers, apparently referring to cloud-based AI services.

The demonstration follows Underdog’s release of Bark, a 167 MB speech-to-text model for edge-device transcription and dictation. Founder Sigil Wen said he wanted to build AI systems that operate locally and allow users to retain control of their data. Underdog has also released a 27-billion-parameter model for laptops and is developing a waitlisted locally hosted app positioned as an alternative to Muse and Instinct.