Note: This bibliographic page is archived and will no longer be updated. For an up-to-date list of publications from the Music Technology Group see the Publications list .

Analysis of Ensemble Expressive Performance in String Quartets: a Statistical and Machine Learning Approach

Title Analysis of Ensemble Expressive Performance in String Quartets: a Statistical and Machine Learning Approach
Publication Type PhD Thesis
Year of Publication 2014
University Univesitat Pompeu Fabra
Authors Marchini, M.
Academic Department Department of Information and Communication Technologies
Abstract Computational approaches for modeling expressive music performance have produced systems that emulate human expression, but few steps have been taken in the domain of ensemble performance. Polyphonic expression and inter-dependence among voices are intrinsic features of ensemble performance and need to be incorporated at the very core of the models. For this reason, we proposed a novel methodology for building computational models of ensemble expressive performance by introducing inter-voice contextual attributes (extracted from ensemble scores) and building separate models of each individual performer in the ensemble. We focused our study on string quartets and recorded a corpus of performances both in ensemble and solo conditions employing multi-track recording and bowing motion acquisition techniques. From the acquired data we extracted bowed-instrument-specific expression parameters performed by each musician. As a preliminary step, we investigated over the difference between solo and ensemble from a statistical point of view and show that the introduced inter-voice contextual attributes and extracted expression are statistically sound. In a further step, we build models of expression by training machine-learning algorithms on the collected data. As a result, the introduced inter-voice contextual attributes improved the prediction of the expression parameters. Furthermore, results on attribute selection show that the models trained on ensemble recordings took more advantage of inter-voice contextual attributes than those trained on solo recordings. The obtained results show that the introduced methodology can have applications in the analysis of collaboration among musicians.
Final publication