Underdog Woof runs a 31.7 MB speech model locally in the browser
Underdog released the open-weights Woof text-to-speech model with an interactive demo that runs locally and offline in the browser through WebGPU.
AT A GLANCE
Key points
- Woof's reported download size is 31.7 MB.
- The Escape demo generates speech locally in the browser through WebGPU.
- The weights are offered under the noncommercial CC BY-NC 4.0 license.
What is changing?
According to Underdog, Woof has 30.7 million parameters and a 31.7 MB download size. In the Escape demo, speech generation happens inside the user's browser on the local GPU through WebGPU, without sending text or audio to the cloud. The provider also says the model runs 18 times faster than real time on a single laptop CPU thread and offers the weights under a CC BY-NC 4.0 license.
The small size could bring low-latency, offline speech generation to desktop and mobile experiences. The speed and quality figures, however, are provider claims rather than independent benchmarks. Commercial use is not directly open under the stated license; teams still need to measure language coverage, voices, device performance and energy use on their own targets and verify licensing scope.
Explore the original sources
Source date: 5 November 2026.
Published by AI Radar: 6 Nov 2026, 17:09 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…