Speech segmentation and word discovery: a computational perspective