Comprehensive and Complex Modeling of Structural Understanding, Studied of an Experimental Improvisation

Music perception and cognition are ruled by complex interdependencies between bottom-up and top-down processes at various cognitive levels, which have not been fully understood and described yet. Cognitive and computational descriptions of particular facets of music listening remain insufficient if they are not integrated in a comprehensive modeling. In the long term, we aim at proposing a comprehensive and complex cognitive modeling of the emergence of structures in music listening and to test its potential by running a computational implementation on elaborate music. The study presented in this paper is part of a broader project, whose general aim is to collect an experimentally controlled jazz improvisation with the view to study jazz listeners’ understanding of that piece. An eminent jazz guitarist, Teemu Viinikainen, was invited to play an original improvisation while following a few general heuristics that we defined beforehand, concerning the use of pauses, repetitions, accentuations and of various ways of evolving the modal discourse. During a subsequent interview, while listening progressively to the recording, the musician gave a detailed a posteriori analysis that was recorded as well, talking and playing examples on his guitar. A systematic analysis was performed exhaustively on the piece, starting from a manual transcription of the piece, followed by motivic, harmonic, rhythmical and structural analyses. Our previous cognitive complex modeling of structural analysis of music has been extended further and implemented in the Matlab programming environment. This extended model starts from the audio recordings, and performs altogether transcription and higher-level analyses, with bottom-up and top-down interactions between low-level and high-level processes. The study challenges the traditional dichotomy between transcription and structural analysis and suggests instead a multi-layer structuring of events of various scales (notes, gestures, motifs, chords, phrases, etc.), where higher-level structures contextually guide the progressive discovery of lower-level elements. The model will be further validated and enriched through a comparison with the musician’s analysis and with jazz listeners’ annotation of the piece collected experimentally.