Aural-Based Detection and Assessment of Real Versus Artificially Synchronized String Quartet Performance

Abstract

In a musical ensemble musicians can influence each other’s performance in terms not only of timing but also in other aspects of the performance such as dynamics, intonation, and timbre. The goal of this work is to test whether this influence can be perceived by a listener from an audio recording solely. We utilize a set of string quartet recordings where every piece is recorded in two experimental conditions: the solo condition, where each musician performs alone; and the ensemble condition, where the musicians perform together after a brief rehearsal. Using state-of-the-art audio analysis/synthesis methods, we artificially synchronize the record-ings in the solo condition note-by-note, thus generating a set of pseudo-ensemble performances where there is no interaction between the musicians. We then carry out a series of listening tests: first, the subjects are tasked with comparing the quality of the performance and the degree of coordination for the two recordings, without knowing that one of them is artificially synchronized. Then, we reveal to the listeners that one of the two versions is artificially synchronized and ask them to point out which recording is which. The results sug-gest that listeners cannot easily discriminate between the real and artificially synchronized recordings; fur-thermore, the accuracy of their judgements appears to be affected by the listeners' level of musical training as well as the piece that is performed.