The speech recognition is a bit spotty, and I wouldn't trust a third party with this on all the time, but it's pretty neat and really fun to see excerpts from what I'm saying and stats on my vocabulary.
Yeah. The speech recognition is really dependent on the placement of your mic, background noise, etc. We're using wit.ai's recognition engine for this, but we could also base it on other libraries or APIs. Perhaps we can figure out a way to optimise the detection in the future.