Misty does person detection using both her time of flight (ToF) sensors and the visual camera. This recognizes not only a person has been found, but they are standing on the dot in front of the robot in the proper spot, all before she starts asking questions. She only records after she has asked a question.
The way the recording works is she’ll ask the screening question and record audio until she hears the end of the speech. This is then sent to Microsoft’s Azure Cognitive Services speech APIs to understand the ‘intent’ (e.g. the APIs process it, not humans) and then the only thing we log is that intent (e.g. yes or no). The total amount we’d record at any point is only 6 seconds.
While Misty is recording, a blue light will show on her side indicating recording is in progress. If that light isn't on, she's not recording.