Underdog Launches 32MB Speech Model for Edge Devices

Underdog has unveiled a new speech model called Woof, which is only 32MB in size and designed to run on edge devices like smartphones and laptops without needing an internet connection. This model is part of a broader movement towards smaller AI models that can operate locally, allowing users to maintain control over their data and reduce dependence on cloud services. The model features 31 million parameters and is available for public download, making it accessible for developers and hobbyists alike.

The Woof model is showcased in a retro-themed game that allows users to interact with various elements, such as a typewriter that speaks back whatever is typed. This demo highlights the potential for local assistant experiences and personalized applications, such as podcasts. Sigil Wen, the founder of Underdog, emphasized the importance of creating AI models that are open and American-made, aiming to empower users with private AI solutions.

In addition to Woof, Underdog has previously launched a 167MB speech-to-text model named Bark, which is also designed for edge devices. The company has plans for a 27-billion-parameter model that can run on laptops, further expanding the capabilities of local AI applications. This shift towards smaller, efficient models could significantly impact how developers approach AI integration in their products, enabling innovative solutions that prioritize user privacy and accessibility.