Segmentation of Tunisian modal improvisation :
Résumé
This study is aimed at an exploration of segmentation strategies through a comparison between listeners’ reactions and predictions estimated using computational models. In a listening experiment, Tunisian subjects of various degrees of expertise were asked to listen to a traditional Tunisian improvisation and to indicate in real time the perceived segmentation, first on a global level and, then, on a more detailed level. In parallel, the same piece has been transcribed and analysed by computer, based on heuristics of local discontinuity and parallelism. A detailed analysis of the possible mapping between listeners’ responses and models’ predictions suggests an explanation of the factors underlying the listeners’ understanding of modal improvisation. Most of the segmentation positions proposed by the subjects can be partly explained by the presence of local discontinuities along time and pitch dimensions. Strong local discontinuities relate to listeners’ segmentation decisions; weaker discontinuities, on the contrary, cannot explain the perception of segmentation unless they are combined with other factors such as parallelism. Segmentation by Tunisian listeners can be approximated using a model combining local discontinuity and parallelism.