MusicXML: An XML Based Approach to Automatic Musicological Analysis

Authorship
  1. 1. Raffaele Viglianti

    Università di Pisa

Work text
This plain text was ingested for the purpose of full-text search, not to preserve original formatting or readability. For the most complete copy, refer to the original conference program.

1. Introduction
Computational music is a young discipline undergoing
ever increasing development. Most research conducted
to date concerns digital audio, multimedia, automatic
composition, score writing, etc. A humanities-oriented approach
(i.e. musicological) is a more recent, less well-defined
development, though it is also growing.
The application of such studies regard, to mention only a few:
• automatic extraction of statistics;
• verification of grammar rules applied to music;
• automatic application of models for formal and harmonic
analysis;
• validation of models for different analysis types.
The MIDI format is the technology most often used in
computational music. Developed in the early 1980s, MIDI is
a digital representation of music, used especially for
communication between electronic musical instruments,
including computers. The MIDI message associates an integer
number to each note (e.g. central C is 60), so it is suitable to a
wide range of studies.
A newer technology, MusicXML, offers the same possibilities
and more, but requires different techniques for data retrieval.
The following abstract presents the results of a statistical
analysis of the tenor parts of some operas by Giacomo Puccini,
encoded in MusicXML for the Research Institution “Centro
Studi Giacomo Puccini” in Lucca, Italy. Specifically, it
examines the arias of Rodolfo from La Bohème and Principe
Calaf from Turandot.
The encoding of these musical pieces was discussed in one of
the author’s thesis dissertation for his degree in Computer
Science for the Humanities at the University of Pisa, under the
supervision of Professor Elena Pierazzo. The dissertation deals
with the possibilities for automatic applications illustrated in
section 4 herein.
2. What is MusicXML?
LLC is an American Internet music publishing and software
company; it developed and promoted the MusicXML
encoding language, which can represent the music notation
system established in the West since the 17th Century.
MusicXML is a royalty-free format that implements all the
features offered by XML technology:
• data structure
• modularity
• extensibilty
• possibility of querying and interaction through XML family
technologies.
MusicXML is designed to represent a score in a digital format,
so most of information contained in a human-readable score is
retrievable in the encoding and can be used for various kinds
of analyses. Features such as lyrics, author or reviewer notes,
dynamics, agogic indications, and so on, are encoded with their
position in the score.
An important part of the encoding is the Score Header, which
contains metadata, such as author, poet, work number,
instruments, etc. An instrument data section even enables
specifying the MIDI instrument channel to be used to play the
music.
3. Automatic statistical extraction:
an example application.
The encoded tenor parts have been inserted into eXist, an
Open Source XML database, to be queried with the
XQuery language.
The goal was to extract statistical data with relevance to the
humanities. Indeed, such data was used for a musicological
dissertation on vocalism in Puccini’s works.
The statistics are based on the criteria proposed by Marco
Gilardone and Franco Fussi in Le voci di Puccini. Un’indagine
sul canto (The Voices of Puccini. An inquiry on singing). The
statistics presented in such work were extended and adapted to
include the ossias.1
The statistics are the percentage occurrences in the distribution
of notes according to the following classification of tones: The upper and lower limits are the extreme notes of the analyzed
part; the numbers beside to the notes denote the octave; the
notes highlighted in grey are transition tones, which are
interesting due to their difficulty of execution within the part.
The original statistics were extended to account for the
incidence of dynamic aspects such as piano, pianissimo,
mezzoforte, forte in transition tones and acute tones.
The XQuery query language was used, as it offers many
possibilities for data extraction from an XML database, though
it is often redundant and does not provide for complex structures
such as arrays or the like.
Starting with the retrieved numerical data, a musicologist can
construct a dramatic character profile. For example, Calaf’s
singing parts exhibit a high percentage of acute tones (27.39%),
transition tones (17.63%) and a high ratio between acute and
grave tones. This is a deliberate aesthetic choice, made to
portray a character in less realistic terms than other tenor roles,
perhaps because he is part of a fantastic drama; indeed, his
vocality tends to be heightened.
Rodolfo, instead, exhibits ‘milder’ musical behavior; he sings
less transition notes (10.37%), and those with dynamic
indications are sung in piano. The middle tones are greatly
predominant (69.63%); Rodolfo is indeed a character who tends
to be more intimate, and displays less dramatic presence.
All the statistics are produced via a query, and each piece of
information needed for calculations is furnished directly from
the encoded score.
Beginning with this model, many other searches may be carried
out, such as graphs or division of the notes in each act, etc.
Deeper analysis with the model provides for clearer delineation
of a character’s profile.
4. Thematic extraction based on
Retì’s musicological analysis
In the 1950s, Rudolf Retì developed a musicological analysis
based on the repetition of thematic cells. Such cells may
occur in different ways and can be found by considering the
melodic interval.2
Consider, for example, the following sequence of notes:
Figure 1
Between C and E there is an interval of two tones, while E and
D are separated by only one tone. If the following sequence
Figure 2
occurs later on in the part, it can be compared to the sequence
in figure 1, because the intervals correspond: there is an interval
of two tones between G and B, and one tone between B and A.
Retì views entire compositions as based on a few thematic cells.
The cell can occur in the same sequence as the initial one or in
a different sequence, that is:
• notes in inverse order
• exchanged position of the notes
• transposition to different notes, but with the same interval.
Any transformation of the cell can also occur with
“interpolated” notes, which are inserted between the principal
notes.
Starting from the corpus encoded in the author’s dissertation
for the Research Institute Centro Puccini, an algorithm was
envisioned for automatic retrieval of the repercussions of such
thematic cells.
Although such algorithm was implemented with XQuery, it
proved unable to fulfill the task, because of the lack of advanced
structures in XQuery. The study is still ongoing, as is the
implementation of the algorithm in other programming
languages, in particular, Java or Perl. 5. Limits of MusicXML
Even if MusicXML offers many possibilities for research,
not every prerequisite of high level academic research
can be satisfied: it is designed to be a sufficient but not optimal
representation of musical scores. For instance, with MusicXML
it is impossible to encode ossias, but for this research it was
enough to encode only one of the two options of the score. On
the other hand , the encoding model would not be optimal for
a study that has as a goal to represent a score exactly as it
appears. Anyway MusicXML is extensible and many other
encoding models can be developed starting from the actual
model. At this stage it should be pointed out that there are other
encoding models that are starting to be used in the academic
research, such as MEI (Music Encoding Initiative). MEI is
designed to represent with effectiveness also information about
the physical support and provides a more complex structure of
metadata. Still the weakness of MEI is the lack of plug-ins that
can convert from score writing software or from optical
recognition programs. As a matter of fact, most research based
on MEI uses XSLT technology to convert from MusicXML to
MEI. This area of research is constantly evolving.
6. Conclusions
The abstract presents MusicXML as an alternative
technology for conducting automatic analysis of music,
adopting a statistical approach. The potentials of such
techniques are considerable: in the future they are likely to be
applied for advanced research, such as pattern matching and
automatic formal analysis, or for added encoding, including
commentaries, metadata for retrieval, harmonic analyses, as
well as for retrieval from digital collections of scores.
1. An ossìa is an alternative way to execute part of a measure or one
or more measures.
2. A melodic interval is the tonal distance between two notes in
sequence.

If this content appears in violation of your intellectual property rights, or you see errors or omissions, please reach out to Scott B. Weingart to discuss removing or amending the materials.

Conference Info

Complete

ADHO - 2007

Hosted at University of Illinois, Urbana-Champaign

Urbana-Champaign, Illinois, United States

June 2, 2007 - June 8, 2007

106 works by 213 authors indexed

Series: ADHO (2)

Organizers: ADHO

Tags
  • Keywords: None
  • Language: English
  • Topics: None