Clustering Symbolic Music Using Paradigmatic and Surface Level Analyses

In this paper, we describe a novel automatic cluster analysis method for symbolic music. The method contains both a surface level and a paradigmatic level analysing block and works in two phases. In the first phase, each music document of a collection is analysed separately: They are first divided into phrases that are consequently fed on a harmonic analyser. The paradigmatic structure of a given music document is achieved comparing both the melodic and the harmonic similarities among its phrases. In the second phase, the collection of music documents is clustered on the ground of their paradigmatic structures and surface levels. Our experimental results show that the novel method finds some interesting, underlying similarities that cannot be found using only surface level analysis.