The Dr. Dobb's website has a 3-minute MP3 that mentions NXT. The coverage starts at the 1:50 mark. It doesn't really mention anything new, but for the first time I caught the statement about the sound sensor responding to 'tones'.
Voice controlled devices have many drawbacks. Not everyone's voice has the same pitch/tone. Volume is an issue, as some people speak louder, some softer. Many voice-controlled or activated devices have to be programmed for a particular person. If someone else's voice doesn't match up... no luck in operating the device.
While voice control might be tricky, if the sound sensor can be programmed to distinguish tones from the high to low spectrum, it opens up the possibility of controlling a robot through many different household item that can project easily identifiable (AND DISTINCT) tones: piano keys, touch-tone phone, guitar strings, etc.
This should be a LOT of fun to play around with...