I do love that this design could only work if the speaker has clear diction, clear logical topics (and train of thoughts), and all while leaving a clear time and space for the AI to respond.
Then they expected it to work with children... have they met children?