Underdog Releases 32MB 'Woof' Speech Model for Edge Device Processing
Startup Underdog has unveiled 'Woof,' a highly compact text-to-speech model measuring only 31.7 MB. Designed to run locally in web browsers and on edge devices without an internet connection, the 31-million-parameter model aims to provide private, low-latency AI experiences such as local assistants or personalized podcasts. The release follows Underdog's recent launch of 'Bark,' a 167 MB speech-to-text model. Founder Sigil Wen emphasizes a mission centered on data privacy and device autonomy, positioning the company's open models as alternatives to cloud-dependent AI providers. In addition to these small-scale models, the company has released a larger 27-billion-parameter model capable of running on standard laptops.
Topics: Edge AI Product Launch Small Language Models Text-to-Speech