dynamically detect the user has stopped speaking

[JohnnySn0w]

current implmentation leaves much to be desired. Short invocations take too long to get a response, and if we made the listen time shorter, longer invocations get cut off before a user finishes speaking.