OFF - LINE AUTOMATie RECOGNITION OF BASIC FO CURVES

A short Polish phrase was spoken by aphonetician with 8 different intonations, and these Prototypes were imitated 10 times each by 15 speakers. Each of the 1200 u t terances was represen ted as a -vector in an 8-D space. 8 quadratic and 8 linear discriminant functions (DFs) were calculated for each of the 1200 utterance-vectors. The DF with the highest value indica ted the assignment of the token to one of the 8 possible classes (types), which was the "recogni tion" of the token, -and the successi ve values mirrored the degree of i ts dissimilari ty from the "type". The representations of -the 8 "types" were assigned according to expectation between 78 I. and 92 I. of the time, and the individual speakers performed between 74 I. and 100 -1. correctly. f!.,alls were found similar to Rise-Falls and there was similari ty among the Rises andbet ween the t wo Rise-Falls.