Biometric Techniques for Classifying Pianists

Previous

Students

MarkDavid Hosale, Media Arts & Tech
Justin Muncaster, Computer Science
Bhaskar Rao, Media Arts & Tech
Max Wiedmann, Undergraduate Researcher

Faculty Advisors

Matthew Turk, Computer Science
Stephen Pope, Media Arts & Tech

Next

Abstract

Staff notation is rich in its ability to communicate to a performer the necessary details to reproduce a musical piece. Pianists can greatly improve their ability simply by reading music and playing pieces. However, more subtle stylistic skills can not be notated and must be interpreted from a given piece. The ability to effectively interpret a piece is what separates the great musicians from the mediocre. It is not clear how to precisely define the stylistic idiosyncrasies of elite pianists. However, it is clear is that students would benefit greatly from mimicking such pianists.

Biometrics deals with the classification of individuals based on biological or behavioral characteristics. Keystroke dynamics, or the analysis of one's typing habits, is one particular biometric that has gained some attention due to its ability to passively classify an individual continuously in real-time. Although keystroke dynamics exhibit the potential for continuous classification, most systems focus on one-time verification or user identity. Developing techniques for continuous identification user would benefit this field.

Our research meets in the middle of these two areas. We wish to develop techniques to identify pianists based on stylistic features extracted as they play a piece. As stylistic features greatly resemble the features one find's in a keystroke dynamics problem, work in one area benefits work in the other. We wish to have a system that can classify pianists in order to allow an apprentice pianist to see how closely he matches the style of his mentor as well as provide algorithms to exploit the potential of keystroke dynamics as a continuous biometric.