Hacker News new | ask | show | jobs
by serf 1287 days ago
run voice through 3 models, when one model is different than the other two then throw it away.

problems abound with that approach, but I could envision it.