Mathematical aspects of the classification of basic pitch patterns

TL;DRAbstract

A method o:E describing, analyzing and classi:Eying :fundamental :Erequency courses in speech is presented. For the analysis of variability of the Fo parameter, the Karhunen -Loeve transformation was used. In order to study the differences between the curves, a discriminant analysis was employed. The resul ts of an a u toma tic analysis demonstra ted the possibili ty of describing time-variable Fo as representing the following typical intonations: Low Rise, High Rise, Full Rise, Low Fall, Full Fall, Level, Low Rise-Fall and Full Rise-Fall in a system of 3 Coordinates. A deterministic classification algorithm was developed. The training set incl uded F o curves which had been judged to be correct imitations of prototype intonations in perceptual tests. The test set consisted of 360 imi ta tions randomly selected from a collection of 1200 and 80/. correct classification was obtained.

Chat with Paper

AI Agents for this Paper