Timbral qualities of semantic structures of music

Abstract

The rapid expansion of social media in music has provided the field with impressive datasets that offer insights into the semantic structures underlying everyday uses and classification of music. We hypothesize that the organization of these structures are rather directly linked with the ”qualia” of the music as sound. To explore the ways in which these structures are connected with the qualities of sounds, a semantic space was extracted from a large collection of musical tags with latent semantic and cluster analysis. The perceptual and musical properties of 19 clusters were investigated by a similarity rating task that used spliced musical excerpts representing each cluster. The resulting perceptual space denoting the clusters correlated high with selected acoustical features extracted from the stimuli. The first dimension related to the high-frequency energy content, the second to the regularity of the spectrum, and the third to the fluctuations within the spectrum. These findings imply that meaningful organization of music may be derived from low-level descriptions of the excerpts. Novel links with the functions of music embedded into the tagging information included within the social media are proposed.