[alxbrun]

True, but in the case of Glass the speaker is always the same (it's the owner) so you can learn and adapt the acoustic model over time. I'm wondering btw if Google doesn't do that already for Android's built-in speech recognition, which accuracy is amazing.

The other problem is that apps cannot (at least for now) change the language model, so Glass will always be in either "search" or "dictation" mode.