For these cases specifically, we use machine learning. There are many cases.
j previous speech k next speech