Discovering an optimal set of minimally contrasting acoustic speech units: a point of focus for whole-word pattern matching.