Dynamic Time Warping for Automatic Musical Form Identification in Symbolical Music Files

摘要

Music information retrieval techniques are used to automatically extract structural data of a piece, however there have been few attempts to study ways to automatically identify the musical form of digital files. In this work we present an implementation of the dynamic time warping algorithm for the automatic identification of musical form structure by means of a segmentation matrix in which we group elements according to maximal similarity. The system was implemented in symbolic files parsed with the music21 library. We tested it in two pieces: Bagatelle No. 25 in A minor by L.V. Beethoven, and Piano Sonata No. 11 in A major, K331, movement 3 by W.A. Mozart. The system obtained a correct identification of the similar sections, both with a rondo form. We foresee that this algorithm can be extended to measure harmonic similarity and with this be able to analyze more complex forms, like a sonata.