A dare · three real voices · all running local
All three are AI voices that run entirely on local hardware — no cloud, no subscription. One is what local TTS sounded like last year. One is a stock voice blend, untouched. One is Boone — the same 82-million-parameter model as the stock blend, after one evening of tuning by ear. He runs on two CPU cores. Listen, then pick him.
The gap between the stock blend and Boone is one evening of listening — a recipe, a pace, two EQ moves, and measured pauses. Every number is written down. That's the method we sell for the price of lunch, and the voice ships in the box we build.
The tuning method → The box he lives in →Fair notes: no humans were imitated — all three are synthetic voices, and Boone's blend is built only from openly licensed stock voices (our rule: we never clone a voice we don't own). Clips render nothing server-side; this page phones nobody. More dares: can you catch the vision model lying? · field notes