|
Elaine Chew: lots of challenges
|
Music expresses and elicits emotion. But how, exactly?
Philosophers have been fascinated by the question since Pythagoras. At
the USC Viterbi School of Engineering, Elaine Chew, assistant professor
of industrial and systems engineering, now teaches a graduate course on
using computational and other engineering tools to look for answers.
Chew, who continues a career as a distinguished concert pianist in
addition to her engineering research, has written an account of the
issues in creating the course in an article that will be presented at
an engineering conference later this year.
Meanwhile the students in "Computational Modeling of Expressive
Performance" presented their results May 3, a collection of 10 projects
now up on the class website as what Chew calls a "non-peer reviewed
publication" at
http://www-scf.usc.edu/~ise575/b/projects
The projects from the class, offered for only the fourth time this
Spring, range widely. Arpi Mardirossian found a trove of silent movie
scores specially written to evoke specific emotions, and analyzed their
characteristics.
|
Analysis of Dynamic Shaping in Unaccompanied Bach (click on images to visit student sites)
|
For "Analysis of Dynamic Shaping in
Unaccompanied Bach," student Eric Cheng created intricate graphs
comparing, note-by-note, performances of a Bach unaccompanied violin
sonata by three master violinists: Jascha Heifetz, Yehudi Menuhin, and
Nathan Milstein, with respect to their dynamic shaping, i.e. the nature
and amount of sound intensity variation.
Meghen Miles and Merrick Mosst created "Emotiongrams" by mapping
specific musical characteristics (e,g,, minor keys), widely identified
with certain emotions (e.g., sadness), to color patterns that
represented varying combinations of energy
|
Emotiongram: Visualization of Emotional Content from Audio
|
and stress.
The other seven topics pose equally interdisciplinary questions -
questions that are quite new in engineering in general, and engineering
education in particular.
In a recently accepted conference publication that will be presented in
fall in San Diego, "A case study in course design at the intersection
of music and engineering," Chew details the challenges in creating the
class, and discusses the problems that remain.
Chew traces the beginning of modern efforts to bring engineering
techniques into analysis of music to the work of Christopher
Longuet-Huggins, a noted theoretical chemist and cognitive scientist,
and gifted amateur musician, who in his "Letters to a Musical Friend"
(The Musical Review, 1962) described computing methods for
"Interpreting Bach" that were implemented and published in Machine
Intelligence in 1971.
|
Clustering of Expressive Music Performances
|
While interest in the field has
mushroomed in the 21st century, with engineering conferences and
refereed journals now covering the subject, teaching is only beginning.
"The challenges include the lack of a formal body of knowledge, in the
form of a text, the lack for formal academic structures to support the
course, the lack of students with suitably strong backgrounds in both
computing and music, and misconceptions about the nature of music
research," she notes in her presentation.
Attacking problems in computational modeling of music "draws upon
methodologies and tools from music theory, cognitive science,
artificial intelligence, experimental psychology, mathematics, signal
processing and neuroscience. Few if any students enrolled in the
course are equipped with the knowledge to understand all the material."
The syllabus devised offers a crash course in the elements of all these
disciplines touching on music cognition, with the aim that, at the end
of the course, each student should be able to "understand basic music
structures; be capable of manipulating digital music; be able to
generate computational means of analyzing, generating and visualizing
structured music; and be able to formulate a question and build the
computational tools to answer it."
|
The Sounds of Silents: - Analysis of Silent Movie Music |
The
course's final projects depend on the students having acquired at least
the rudiments of all these skills. The amount of material that has to
be mastered is so great that Chew reluctantly had to move what had been
a popular feature of the course - guest appearances by musicians and
researchers in the field - to its own separate series.
Precisely because of the unusual mixture of disciplines involved,
making this course a reality was not institutionally easy, Chew
says. However, a remarkable record of recognition for her
research (she holds a Ph.D. from M.I.T in engineering, and a FTCL from
Trinity College, London in piano performance, an Early Career / PECASE
award from the National Science Foundation, and the Viterbi Early
Career Chair from USC) helped to pave the way.
"All these interdisciplinary education and outreach activities could not have happened without the staunch support of my own
|
Following a Pianist's Pace Exploring the Information Revealed by Tempo Changes
|
department, the deans of the Viterbi
School, and the provost," Chew says. "My own department has
graciously allowed me to create this special topics class in lieu of
teaching another traditional industrial and systems engineering
course. My department chair, James Moore, has encouraged me to
forge ahead in creating an undergraduate counterpart to the class
because he sees the potential of such courses in recruitment and
retention of young and inquiring minds to engineering."
Says Moore, chair of the Daniel Epstein Department of Industrial and
Systems Engineering, her professional home: "Taking interdisciplinary
work seriously involves accepting some intellectual risks. Her work is
novel, but important; and this makes her course novel and
important. It is a source of no small pride for us that the most
appropriate avenue for connecting Prof. Chew, her students, and her
work to the USC Viterbi School is the Daniel J. Epstein ISE Department.
|
Tempo and Loudness Tracking via Frequency and Time Domain Analysis for Polyphonic Music
|
With this kind of class, it's not just a
matter of "if you build it, they will come," said Chew:
recruiting students was a major effort. "I personally and actively
publicized the class widely through emails and posters to other
departments in the Viterbi (Electrical Engineering, Computer Science,
Bioengineering) and Thornton schools (Composition, History and
Literature, Education), and the College (Psychology, Neuroscience,
Mathematics)."
What kind of student winds up in a class like this? "This year, the
class consisted of students from ISE, EE, CS, and Neuroscience; we had
a vocalist, a violinist, a pianist, some with a couple years of
instrumental training, and a few with no musical background."
For example, Meghen Miles: "I have a Bachelors' degree from the
Thornton School of Music, and now I'm getting a Ph.D. in Neuroscience.
I found out about the class from emails from quite a few sources,
including Elaine Chew herself. I thought the class would help me both
with my understanding of computational modeling and by introducing me
to a new body of literature with direct ties to my research.
Arpi Mardirossian is a Ph.D. candidate in ISE. "I walk away from
the class knowing much more about music expression and its state of the
art."
So: how does music express and evoke emotion? Elaine Chew and her students don't know yet. But check back in a few years.
Elaine Chew's research was supported by the National Science
Foundation. Her presentation will appear at the 36th ASEE/IEEE
Frontiers in Education Conference, October 28-31, in San Diego
CA. This story was also widely published on the web, including
the following:
