Underdog shows off a 32MB speech model in new demo
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
A 31M-parameter, 31.7MB text-to-speech model now runs entirely in-browser with no internet connection, demonstrated via a playable WebAssembly-style game. This puts fully local, zero-latency, zero-cost TTS within reach for client-side assistants, dictation, and podcast generation—no cloud API bills or round-trips—though quality won't match larger hosted voices, so treat it as an edge/offline fallback or privacy-first tier rather than a drop-in replacement for your current voice stack.
Underdog has released "Woof," a 31.7 MB text-to-speech model that runs entirely on-device and in-browser without an internet connection, complementing their 167 MB "Bark" speech-to-text model. This allows you to deploy fully local, zero-latency voice-to-voice agents directly on client edge devices, completely eliminating server-side GPU hosting costs and network latency for your conversational audio pipelines.