Classification of basic fundamental frequency patterns using discriminant functions

An attempt was made to see whether realizations of eight different pitch patterns can be classified automatically by subjecting fundamental frequency curves to statistical treatment using two kinds of discriminant functions. The materials consisted of 150 imitations of each of the eight prototypes (eight intonations of a short Polish phrase). Approximately 80% of the time the statistical procedures assigned the individual tokens to the appropriate classes. Most of the misassignments were due to wrong imitations. The discriminant functions also indicated similarities and dissimilarities among the pitch patterns. It is suggested that perceptually and linguistically, some differences between pitch patterns may not be quite categorical.