Assessing A Music Student's Progress

Teachers frequently make errors when assessing music students. We propose a machine learning application that, given two performances of a piece of music, determines which performance is better, providing an objective and accurate assessment of progress. Several features are extracted from performances using music analysis algorithms, creating a vector of features for each performance. The vectors from two performances of a piece of music are subtracted from each other, and this vector of differences is input to a machine learning classifier which maps the vector to an assessment of progress. The implementation demonstrates that such a tool is feasible.