Not All Roads Lead to Rome: Pitch Representation and Model Architecture for Automatic Harmonic Analysis. Micchi, G., Gotham, M., & Giraud, M. Transactions of the International Society for Music Information Retrieval, 3(1):42–54, 2020. doi abstract bibtex Automatic harmonic analysis has been an enduring focus of the MIR community, and has enjoyed a particularly vigorous revival of interest in the machine-learning age. We focus here on the specific case of Roman numeral analysis which, by virtue of requiring key/functional information in addition to chords, may be viewed as an acutely challenging use case. We report on three main developments. First, we provide a new meta-corpus bringing together all existing Roman numeral analysis datasets; this offers greater scale and diversity, not only of the music represented, but also of human analytical viewpoints. Second, we examine best practices in the encoding of pitch, time, and harmony for machine learning tasks. The main contribution here is the introduction of full pitch spelling to such a system, an absolute must for the comprehensive study of musical harmony. Third, we devised and tested several neural network architectures and compared their relative accuracy. In the best-performing of these models, convolutional layers gather the local information needed to analyse the chord at a given moment while a recurrent part learns longer-range harmonic progressions. Altogether, our best representation and architecture produce a small but significant improvement on overall accuracy while simultaneously integrating full pitch spelling. This enables the system to retain important information from the musical sources and provide more meaningful predictions for any new input.
@Article{ micchi.ea2020-not,
author = {Micchi, Gianluca and Gotham, Mark and Giraud, Mathieu},
year = {2020},
title = {Not All Roads Lead to Rome: Pitch Representation and
Model Architecture for Automatic Harmonic Analysis},
abstract = {Automatic harmonic analysis has been an enduring focus of
the MIR community, and has enjoyed a particularly vigorous
revival of interest in the machine-learning age. We focus
here on the specific case of Roman numeral analysis which,
by virtue of requiring key/functional information in
addition to chords, may be viewed as an acutely
challenging use case. We report on three main
developments. First, we provide a new meta-corpus bringing
together all existing Roman numeral analysis datasets;
this offers greater scale and diversity, not only of the
music represented, but also of human analytical
viewpoints. Second, we examine best practices in the
encoding of pitch, time, and harmony for machine learning
tasks. The main contribution here is the introduction of
full pitch spelling to such a system, an absolute must for
the comprehensive study of musical harmony. Third, we
devised and tested several neural network architectures
and compared their relative accuracy. In the
best-performing of these models, convolutional layers
gather the local information needed to analyse the chord
at a given moment while a recurrent part learns
longer-range harmonic progressions. Altogether, our best
representation and architecture produce a small but
significant improvement on overall accuracy while
simultaneously integrating full pitch spelling. This
enables the system to retain important information from
the musical sources and provide more meaningful
predictions for any new input.},
doi = {10.5334/tismir.45},
journal = {Transactions of the International Society for Music
Information Retrieval},
keywords = {1,1 key,chords and functional harmony,computational
musicology,corpus,functional harmony,introduction,is
common to a,machine learning,motivation,pitch
encoding,previous work,roman numeral analysis,some sense
of,tonal harmony,very wide},
mendeley-tags= {computational musicology},
number = {1},
pages = {42--54},
volume = {3}
}
Downloads: 0
{"_id":"3guhgR3kpeHra2qyJ","bibbaseid":"micchi-gotham-giraud-notallroadsleadtoromepitchrepresentationandmodelarchitectureforautomaticharmonicanalysis-2020","authorIDs":[],"author_short":["Micchi, G.","Gotham, M.","Giraud, M."],"bibdata":{"bibtype":"article","type":"article","author":[{"propositions":[],"lastnames":["Micchi"],"firstnames":["Gianluca"],"suffixes":[]},{"propositions":[],"lastnames":["Gotham"],"firstnames":["Mark"],"suffixes":[]},{"propositions":[],"lastnames":["Giraud"],"firstnames":["Mathieu"],"suffixes":[]}],"year":"2020","title":"Not All Roads Lead to Rome: Pitch Representation and Model Architecture for Automatic Harmonic Analysis","abstract":"Automatic harmonic analysis has been an enduring focus of the MIR community, and has enjoyed a particularly vigorous revival of interest in the machine-learning age. We focus here on the specific case of Roman numeral analysis which, by virtue of requiring key/functional information in addition to chords, may be viewed as an acutely challenging use case. We report on three main developments. First, we provide a new meta-corpus bringing together all existing Roman numeral analysis datasets; this offers greater scale and diversity, not only of the music represented, but also of human analytical viewpoints. Second, we examine best practices in the encoding of pitch, time, and harmony for machine learning tasks. The main contribution here is the introduction of full pitch spelling to such a system, an absolute must for the comprehensive study of musical harmony. Third, we devised and tested several neural network architectures and compared their relative accuracy. In the best-performing of these models, convolutional layers gather the local information needed to analyse the chord at a given moment while a recurrent part learns longer-range harmonic progressions. Altogether, our best representation and architecture produce a small but significant improvement on overall accuracy while simultaneously integrating full pitch spelling. This enables the system to retain important information from the musical sources and provide more meaningful predictions for any new input.","doi":"10.5334/tismir.45","journal":"Transactions of the International Society for Music Information Retrieval","keywords":"1,1 key,chords and functional harmony,computational musicology,corpus,functional harmony,introduction,is common to a,machine learning,motivation,pitch encoding,previous work,roman numeral analysis,some sense of,tonal harmony,very wide","mendeley-tags":"computational musicology","number":"1","pages":"42–54","volume":"3","bibtex":"@Article{ micchi.ea2020-not,\n author = {Micchi, Gianluca and Gotham, Mark and Giraud, Mathieu},\n year = {2020},\n title = {Not All Roads Lead to Rome: Pitch Representation and\n Model Architecture for Automatic Harmonic Analysis},\n abstract = {Automatic harmonic analysis has been an enduring focus of\n the MIR community, and has enjoyed a particularly vigorous\n revival of interest in the machine-learning age. We focus\n here on the specific case of Roman numeral analysis which,\n by virtue of requiring key/functional information in\n addition to chords, may be viewed as an acutely\n challenging use case. We report on three main\n developments. First, we provide a new meta-corpus bringing\n together all existing Roman numeral analysis datasets;\n this offers greater scale and diversity, not only of the\n music represented, but also of human analytical\n viewpoints. Second, we examine best practices in the\n encoding of pitch, time, and harmony for machine learning\n tasks. The main contribution here is the introduction of\n full pitch spelling to such a system, an absolute must for\n the comprehensive study of musical harmony. Third, we\n devised and tested several neural network architectures\n and compared their relative accuracy. In the\n best-performing of these models, convolutional layers\n gather the local information needed to analyse the chord\n at a given moment while a recurrent part learns\n longer-range harmonic progressions. Altogether, our best\n representation and architecture produce a small but\n significant improvement on overall accuracy while\n simultaneously integrating full pitch spelling. This\n enables the system to retain important information from\n the musical sources and provide more meaningful\n predictions for any new input.},\n doi = {10.5334/tismir.45},\n journal = {Transactions of the International Society for Music\n Information Retrieval},\n keywords = {1,1 key,chords and functional harmony,computational\n musicology,corpus,functional harmony,introduction,is\n common to a,machine learning,motivation,pitch\n encoding,previous work,roman numeral analysis,some sense\n of,tonal harmony,very wide},\n mendeley-tags= {computational musicology},\n number = {1},\n pages = {42--54},\n volume = {3}\n}\n\n","author_short":["Micchi, G.","Gotham, M.","Giraud, M."],"key":"micchi.ea2020-not","id":"micchi.ea2020-not","bibbaseid":"micchi-gotham-giraud-notallroadsleadtoromepitchrepresentationandmodelarchitectureforautomaticharmonicanalysis-2020","role":"author","urls":{},"keyword":["1","1 key","chords and functional harmony","computational musicology","corpus","functional harmony","introduction","is common to a","machine learning","motivation","pitch encoding","previous work","roman numeral analysis","some sense of","tonal harmony","very wide"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"article","biburl":"https://hmb.sampaio.me/bibliografia.bib.txt","creationDate":"2020-11-02T10:30:14.981Z","downloads":0,"keywords":["1","1 key","chords and functional harmony","computational musicology","corpus","functional harmony","introduction","is common to a","machine learning","motivation","pitch encoding","previous work","roman numeral analysis","some sense of","tonal harmony","very wide"],"search_terms":["roads","lead","rome","pitch","representation","model","architecture","automatic","harmonic","analysis","micchi","gotham","giraud"],"title":"Not All Roads Lead to Rome: Pitch Representation and Model Architecture for Automatic Harmonic Analysis","year":2020,"dataSources":["n6MFY2CscQLDpJ7nT"]}