A compact voice model runs directly in the browser
AI startup Underdog has introduced Woof, a 31.7 MB text-to-speech model showcased through an interactive browser game. The demo lets users type text on a virtual typewriter and hear it spoken, while a music tuner produces different sounds.
Underdog says Woof has 31 million parameters and can run on many devices without an internet connection. The model is also available to download. Its small size could make it suitable for on-device tools such as voice assistants and personalized audio experiences, though it is not intended to replace larger models.
The release follows the company's 167 MB Bark speech-to-text model, designed for tasks including transcription and dictation. Underdog is also developing a larger model for laptops and an app built around local AI processing.
By bringing speech generation onto everyday devices, compact models like Woof could help make AI tools more responsive and give users greater control over how their data is handled.
