ABSTRACT

This chapter assesses to what extent human musicians improvising jazz with a shared lead sheet actually interact with each other during solos. It proposes a framework for analysis, based on the comparison of correlation estimators computed when musicians play together and when they do not. The chapter illustrates the approach with the analysis of multi-track audio recordings of jazz performances. It considers the correlation between the solo player and the rhythm section, as well as between the members of the rhythm section. The chapter analyzes the audio signals corresponding to performances where the musicians played concomitantly and when they did not. It argues that either content-based interaction in jazz is a myth or that interactions do take place but at yet unknown musical dimensions. The chapter evaluates the correlations between couples of time-series corresponding to the same song and different instruments. Several methods of correlations are: Pearson's correlation coefficient, cross-correlation functions, Granger causality, windowed mutual information, and information dynamics.