How accurate is the voice recognition? Even when restricted to a few command phrases, I struggle to see how this can be overly accurate when running on a Raspberry Pi. I've played with voice recognition in the past, and anything less than 98% accuracy is super frustrating after a week of usage. Siri is the most accurate voice recognition software I've used so far, and even Siri regularly fails to understand some words.

It depends on the tts plugin you use. We have build and tested several tts plugins. I can recommend Microsoft Bing Voice and the new Google Speech API. Both are very rarely wrong.

Aren't both of those "send the speech to the server" systems?