OpenEars looks to be a very nice open source library that can convert
speech to text and render text to speech. It can work in a background thread
and claims to only use 10% CPU on an iPhone 3G. I haven’t tried it myself but
it could be a great resource for continual speech recognition.
It’s licensed under a modified CDDL v1.0. The exception granted
means you don’t have to release your iOS app’s source if you don’t want to.