Methods for automatic alignment of performances to a score representation

The purpose of this document is to report a method for automatic alignment of audio recordings of musical performances to a score representation of the performed piece. The alignment method has been evaluated experimentally on a dataset of classical piano music, and was found to yield more accurate alignments than Dynamic Time Warping, an alignment method widely used in the literature.

This method plays a key role in several of the problems addressed in the PHENICX project. Specifically, accurate score-performance alignments allow for applications such as displaying sheet music synchronously while playing audio/video of recorded performances, and interactively navigating through recorded audio/video, based on the musical score, or annotations/representations derived from the score, such as a structural analysis of the piece. Furthermore, score-performance alignments are the starting point for extracting expressive parameters from recorded performance, such as tempo and loudness curves (see Deliverable 4.1 for more information). This document is intentionally kept very short, since the essential text describing the method and its evaluation is contained in a peer-reviewed published paper, included under Appendix A. Apart from the goal statement (Section 1.1) and executive summary (Section 1.2), the rest of this document is a brief contextualization of the published work with respect to the deliverable objective (Section 2), and a conclusion, where we point to application areas of the method elsewhere in the PHENICX project, and give some future directions based on this work (Section 3).