Adaptive frequency mapping of speech signals