香蕉视频

The Doctoral Colloquium is open to all.

Doctoral Colloquium:听Timothy de Reuse, PhD Candidate in Music Technology

Title:听鈥淲hat's Wrong with This Music? Detecting Errors in Optical Music Recognition Output using Sequence Alignment and Machine Learning鈥�

Biography: Timothy de Reuse is a PhD Candidate in Music Technology. His research centers around analysis of repeated material in symbolic music; in particular, analysis of long-term dependencies in music (e.g., verse-chorus forms or sonata forms). He has published work on automatic motivic analysis, musical pattern discovery, and the use of repetition to find errors in musical scores.

Abstract: Optical Music Recognition is the field of research investigating digitization of physical musical scores into machine-readable forms, enabling researchers to search, analyze, and interact with vast swathes of musical data in novel ways. However, all forms of this technology introduce non-trivial errors into processed scores, especially on poor-quality scans or degraded historical musical documents, and manual correction of these errors time-consuming. I note that most of the errors these processes induce are "non-musical'' in that they represent phenomena that a human composer would never have written. I propose that the laborious task of manual correction could be sped up by marking all symbols on a score that are musically unlikely, allowing the corrector to focus their attention only on regions of the score that contain errors. This dissertation covers the design, construction, and evaluation of a machine learning-based error detector for symbolic music, built to expedite the process of manual correction of optical music recognition outputs. This research aims to help bridge the gap between the promises of optical music recognition and its current limitations, propelling these technologies to their full potential as tools for musicological research and musical preservation.

听