Analyzing directionality of influence among ensemble musicians using Granger Causality

Abstract

In small musical groups, performers can seem to coordinate their movements almost effortlessly in remarkable exhibits of joint action and entrainment. To achieve a common musical goal, co-performers interact and communicate using non-verbal means such as upper-body movements, and particularly head motion. Studying these phenomena in naturalistic contexts can be challenging since most techniques make use of motion capture technologies that can be intrusive and costly. To investigate an alternative method, we analyze video recordings of a professional instrumental ensemble by extracting trajectory information using pose estimation algorithms. We examine Kansei perspectives such as the analysis of non-verbal expression conveyed by bodily movements and gestures, and test for causal relationships and directed influence between performers using the Granger Causality method. We compute weighted probabilities representing the likelihood that each performer Granger Causes co-performers’ movements. Effects of different aspects of musical textures were examined and results indicated stronger directionality for homophonic textures (clear melodic leader) than polyphonic (ambiguous leadership).