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 .

Computational Modeling of Expressive Music Performance New Machine Learning Approaches for Dealing With Real-World Data

Title Computational Modeling of Expressive Music Performance New Machine Learning Approaches for Dealing With Real-World Data
Publication Type Master Thesis
Year of Publication 2006
Authors Hazan, A.
preprint/postprint document files/publications/49b22b-DEA-2006-Hazan.pdf
Abstract We present a doctoral pre-thesis report focusing of inductive computational modeling of music performance.

We first introduce the expressive performance phenomenon from various points of view musicology, music cognition, computer music and brain science. We then review the approaches for modeling this phenomenon with an emphasis on inductive models i.e. models built from real-world performance data. The process of building an expressive performance database from acoustical recording is explained.

Then, we present a comparison of well-established data mining (DM) techniques for inducing models from this database. This comparison has several implications that show the limits of standard DM techniques for building expressive performance models. We then propose a new approach called Evolutionary Population of Generative Models (EPGM) based on Evolutionary Computation (EC) which is suitable for dealing with the issues that arise when using the standard DM paradigm for performance modeling. We introduce several strategies for biasing the model search towards useful solutions, and propose alternatives for evaluating the accuracy of the evolved models.

We validate the results using the performance database presented above and study the impact of our design decisions. We discuss the potential implications this new method can have in the EC field. Finally, we present some conclusions and dress a proposal which has implications in both expressive music performance modeling, machine learning, and evolutionary computation fields.