Underdog Woof is a 40.6 MB English text-to-speech model with two voices that runs on a laptop CPU or in the browser
Underdog Woof is a small English text-to-speech model with two voices, Juno and Mira, published with open weights on 5 November 2026. The two-voice model is a single 40.6 MB file that runs on a laptop CPU with only numpy, and a 31.7 MB one-voice version runs in the browser at underdog.ai/woof.
What it does
It is built to read phone numbers, dates, prices and codes the way people say them; on the Gradium TTS Hard Cases set the makers measured a 1.42% word error rate for the Mira voice, against 1.89% for the 41.5 GB Fish Audio Drama 3. The Juno voice runs 13.8 times faster than real time on one Apple M5 Max CPU thread with 402 MB of peak memory.
Worth knowing
The weights are under the Underdog Non-Commercial License 1.0, so it is not for paid work without other terms. It speaks English only with two voices, and all benchmark numbers are the makers' own, with transcripts and errors scored by other AI models.