...
For example, suppose the NPC's first line starts with "Hello". If the button is pressed at this point, the PC might simply respond with "Hi! How are you?" If the NPC is not interrupted and gets to continue talking, he/she might say, "Hello. And how is Fluffy doing?", leading the PC to respond with a comment on Fluffy's health. If the NPC still isn't interrupted, the full line might turn out to be, "Hello. And how is Fluffy doing? Nice cat, Fluffy. Too bad if something were to happen to her..." Now the PC's next line would become a reaction to the threat. And so on.
If you're interested, there's also some background on the game idea in this writeup of Marleigh's talk at GDC: http://www.gamespot.com/news/6280781.html
In AudioEditor, the text lines are marked with asterisks to indicate how far into the NPC's line the PC's response changes. Draggable breakpoints serve the same purpose in the audio files. AudioEditor also contains a second tab, which stores the background music, the "end game" message, and the audio prompt played if the user does not press the button by the end of the NPC's line. It will probably become clearer how all this works if you play the sample game a few times in ClickToSpeak and then open the XML file in AudioEditor.
...