I attended a session in which Didier Brown demonstrated a Voice gesture. The basic idea is that you turn the voice recognition problem into a pattern recognition problem. Demo here: http://vimeo.com/8203323
From flash 10.1, you can access microphone data at a low level, which allows the developer to create an audio spectrogram.
Although there are many server-side solutions to this problem, unfortunately, there are not many client-side solutions.
source share