Edinburgh Research Explorer

Evaluating Automatic Polyphonic Music Transcription

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Related Edinburgh Organisations

Open Access permissions

Open

Documents

  • Download as Adobe PDF

    Rights statement: © Andrew McLeod, Mark Steedman. Licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). Attribution: Andrew McLeod, Mark Steedman. “Evaluating Automatic Polyphonic Music Transcription”, 19th International Society for Music Information Retrieval Conference, Paris, France, 2018.

    Final published version, 315 KB, PDF document

    Licence: Creative Commons: Attribution (CC-BY)

http://ismir2018.ircam.fr/doc/pdfs/148_Paper.pdf
http://ismir2018.ircam.fr/pages/events-main-program.html
Original languageEnglish
Title of host publicationProceedings of the 19th International Society for Music Information Retrieval Conference, ISMIR 2018
Subtitle of host publicationParis, France, September 23-27, 2018
EditorsEmilia Gómez, Xiao Hu, Eric Humphrey
Pages42-49
Number of pages8
Publication statusPublished - 20 Nov 2018
Event19th International Society for Music Information Retrieval Conference - Paris, France
Duration: 23 Sep 201827 Sep 2018
http://ismir2018.ircam.fr/

Conference

Conference19th International Society for Music Information Retrieval Conference
Abbreviated titleISMIR 2019
CountryFrance
CityParis
Period23/09/1827/09/18
Internet address

Abstract

Automatic Music Transcription (AMT) is an important task in music information retrieval. Prior work has focused on multiple fundamental frequency estimation (multi-pitch detection), the conversion of an audio signal into a timefrequency representation such as a MIDI file. It is less common to annotate this output with musical features such as voicing information, metrical structure, and harmonic information, though these are important aspects of a complete transcription. Evaluation of these features is most often performed separately and independent of multi-pitch detection; however, these features are non-independent.
We therefore introduce MV 2H, a quantitative, automatic, joint evaluation metric based on musicological principles, and show its effectiveness through the use of specific examples. The metric is modularised in such a way that it can still be used with partially performed annotation— for example, when the transcription process has been applied to some transduced format such as MIDI (which may itself be the result of multi-pitch detection). The code for the evaluation metric described here is available at https://www.github.com/apmcleod/MV2H.

Event

19th International Society for Music Information Retrieval Conference

23/09/1827/09/18

Paris, France

Event: Conference

ID: 81559199