Underdog Launches 32MB Speech Generation Model Woof

Startup Underdog has launched Woof, an open-source speech generation model with just 31 million parameters and a 32MB footprint, showcased through an interactive web demo. The model can run entirely offline directly in the browser and across most edge devices without internet connectivity, converting text input into spoken audio. Underdog previously released Bark, a 167MB on-device speech-to-text model, as part of its push to build a privacy-preserving local AI ecosystem.