How much training data do I need for a CNN that recognizes the vocalizations of a single person?

Question

I have a friend that communicates through unique vocalizations (not in any language), and I want to build a CNN that can recognize these sounds. I know that for speech recognit