Automatic Piano Reduction of Orchestral Music Based on Musical Entropy

In this paper, we present a method for arranging orchestral pieces for piano by evaluating the amount of information contained in musical phrases in order to reduce the musical score to make it playable on piano, while retaining as much information of the original piece as possible. While previous systems for automatic arrangement often rely on heuristic rules, the core of our algorithm is the computation of the entropy of several musical features to detect the most important parts of an orchestral piece. The extracted features were chosen to evaluate the melodic and rhythmic complexity of phrases which is assumed to be related to the importance of these phrases for the character of the original piece. A sliding frame analysis method is used to detect phrase segmentation in melodies moving through different voices. The algorithm finally yields a piano score that contains as many of the important notes as possible, being further constrained by criteria that ensure the playability on a piano. The effectiveness of the piano reduction principle is shown in arrangements generated from Beethoven's symphonies.