> The free version, called Subvox, doesn't read speech muscles. It reads the head-motion data AirPods already collect, then matches your jaw and head movements against words you have trained it on. The instructions suggest five words, ten times each. It has no language model and it publishes no accuracy score.
> So it won't take dictation. One critic said it "would fall apart with open vocabulary," and that sounds right to me.
> The free version, called Subvox, doesn't read speech muscles. It reads the head-motion data AirPods already collect, then matches your jaw and head movements against words you have trained it on. The instructions suggest five words, ten times each. It has no language model and it publishes no accuracy score.
> So it won't take dictation. One critic said it "would fall apart with open vocabulary," and that sounds right to me.