Improving optical music recognition by combining outputs from multiple sources
- Submitting institution
-
The University of Lancaster
- Unit of assessment
- 32 - Art and Design: History, Practice and Theory
- Output identifier
- 236021403
- Type
- E - Conference contribution
- DOI
-
-
- Title of conference / published proceedings
- Proceedings of the 16th International Society for Music Information Retrieval Conference
- First page
- 517
- Volume
- -
- Issue
- -
- ISSN
- -
- Open access status
- -
- Month of publication
- October
- Year of publication
- 2015
- URL
-
http://ismir2015.uma.es/
- Supplementary information
-
-
- Request cross-referral to
- 33 - Music, Drama, Dance, Performing Arts, Film and Screen Studies
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- Yes
- Number of additional authors
-
3
- Research group(s)
-
-
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- This paper describes the outcome of an AHRC-funded project. Optical music recognition (OMR) is the only realistic way of converting the vast quantity of existing musical scores into a format which allows computer software to use the musical information in the scores, rather than treating them simply as images, like the text of an unreadable language. However, the accuracy of available OMR software is too low to be usable for many purposes without substantial and time-consuming human intervention. By combining outputs from multiple sources, the techniques developed in this research are able to reduce the typical number of errors in OMR by about half. This paper is the last in a series which reported the outcome of the project at conferences on the electronic arts (EVA 2014), digital libraries for musicology (DLfM 2014), music analysis (EuroMAC 2014), and music information retrieval (ISMIR 2015), reaching an overall audience of several hundred. Software which provides a configurable pipeline for applying multiple OMR software to multiple sources, and which combines the results to produce the most accurate outcome, is provided in two public repositories. The project involved two researchers at each site, one providing high-level guidance and direction (Marsden at Lancaster and Ng at Leeds) and one doing the coding and conducting the experiments (Padilla at Lancaster and McLean at Leeds). Marsden was the overall PI, originator of the idea, and principal guide for the research. He and Padilla worked closely on the final experiments. The text of the paper was largely written by Marsden, with substantial contributions from Padilla and minor contributions from the other authors. The work has been cited in 21 other papers, several of which use its outcome in making further improvements to the quality of OMR.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -