EXPLORING MELODIC CONTOUR: A CLUSTERING APPROACH

P R E V I O U S S T U D I E S I N V E S T I G A T I N G C O M M O N melodic contour shapes have relied on methodologies that require prior assumptions regarding the expected contour patterns. Here, a new approach for examining contour using dimensionality reduction and unsupervised machine-learning clustering methods is presented. This new methodology was tested across four sets of data: two sets of European folk songs; a mixed-style, curated dataset of Western music; and a set of Chinese folk songs. In general, the results suggest the presence of four broad common contour shapes across datasets: convex, concave, descending, and ascending. In addition, the analysis revealed some micro-contour tendencies, such as pitch stability at the beginning of phrases and descending pitch at phrase endings. These results are in line with previous studies of melodic contour and provide new insights regarding the prevalent contour characteristics in Western music.