It has been shown that listeners are sensitive to pitch distributional information in music (Oram & Cuddy, 1995; Krumhansl, Louhivuori, Toiviainen, Järvinen, & Eerola, 1999). However, it is uncertain whether frequency-based musical features are sufficient in explaining listeners' similarity judgements which underlie listeners' classification processes. A similarity rating experiment was designed to determine the effectiveness of these features in predicting listeners' similarity ratings. The material consisted of 15 melodies representing five folk music styles. A multiple regression analysis showed that the similarity of frequency-based musical properties could account for a moderate amount (40%) of listeners' similarity ratings. A slightly better predictive rate (55%) was achieved using descriptive variables such as number of tones, rhythmic variability and melodic predictability. The results suggest that both measures were able to capture some aspects of the structures which portray common salient dimensions to which listeners pay attention whilst categorizing melodies.
Keyword(s): Music cognition; Categorization; Statistical features; Similarity; Melody;