Well, we'll need a voice to text generator. Then we'll need some kind of AIML handler. Finally, a text to voice generator. IBM use to sell a Voice-To-Text interface card in the late 1990's. Text to Voice is a small software routine these days.
The Machine Learning part is the intriguing part. Books have been, and will continue to be written on this. What the hard part is, "Is how can a computer program find a valid fact, and be able to defend that the fact is valid?"