Graham Percival
Email: <g.percival AT SPAMFREE elec SMALL-ROUND-THING gla SPEC-OF-DUST ac COUNTRY-OF uk>
Home page: http://percival-music.ca
Computer-Assisted Musical Instrument Tutoring
Learning to play a musical instrument is a daunting task. Musicians must execute unusual physical movements within very tight tolerances, and must continually adjust their bodies in response to auditory feedback. However, most beginners lack the ability to accurately evaluate their own sound. We therefore turn to computers to analyze the student's performance. By extracting certain information from the audio, computers can provide accurate and objective feedback to students.
The introduction to my oral thesis defense is a good introduction: camit-talk.pdf
Projects
"Listen like a music teacher": Analysis of musical exercise performances
Software such as MEWsician's Exercises for Rhythms (MEWER) and Musician Evaluation and AWdition for Strings (MEAWS) grade simple exercises. Commercial games such as Guitar Hero, Rock Band, and various Karaoke-singing game provide grades for playing/singing whole pieces of music. However, the musical foundation of these grades is questionable (notoriously so, for the karaoke games!).
This project aims to devise objective grading systems for various aspects of music, backed by real-world experiments to collect the judgements of professional musicians and music teachers. The aim is to create computer software which gives the same grades as a real music teacher.
Various aspects of music
Creating software which provides reasonable grades for performing a piece of music is impossible -- or at least, highly unlikely within my lifespan. Any kind of objective evaluation -- for any reasonable, computer-created grade must be objective -- would require an objective description / definition of musical aesthetics.
However, specific musical skills can be evaluated objectively. For example, there is only one answer to performing a specific rhythm at a given tempo. If we measure a students' rhythmic performance, then we can evaluate their rhythmic accuracy.
Consider this rhythmic performance shown on the right. The black bars indicate where the student should have clapped; the red bars indicate where the student did clap. This information is presented as a list of numbers below:
Expected: 0 86 129 172 215 258 344 430 473 516 559 602 Aligned: 0 86 132 175 219 262 351 421 458 514 560 603 Differences: 0 0 3 3 4 4 7 9 15 2 1 1
The task is to provide a grade based on these comparisons; we must learn about the psychology of rhythmic perception, ideally checking musicians and non-musicians.
Same grades
Each music teacher will assign different grades; we are not looking for a universal agreement that the above exercise deserves 83% or 57% or what have you. Rather, we want to find agreement that exercise X is better than exercise Y. If we find agreement about the relative ranking of exercises, then we can devise an algorithm which ranks exercises in the same order. At that point, we are left with a single variable which provides an overall scaling factor; this may be adjusted based on the age of the child (or in general, the desired difficulty level).
"Read like a music teacher": Analysis of sheet music
Before giving a new piece of music to a student, a music teacher will quickly glance at it to judge the difficulty level. Unless a student is extremely motivated, the teacher will want to choose a piece which is slightly challenging, but not impossibly so.
An automated method of analysing sheet music -- again, by splitting the analysis into various aspects of music, rather than music as a whole -- would be a useful tool for CAMIT.
In addition to the obvious educational benefits, this could be very interesting for analysing scores. String players have a vague notion that classical music is easier than Romantic music, which in turn is easier than modern music. But it would be nice to quantify this difference. In addition, with such automatic tools, we could compare music for other instruments -- does the difficulty of piano or wind music change over the ages? If so, is the change related to technical improvements in instrument construction, or a general artistic desire for more elaboration (or more dramatic gestures, or whatever), etc?
"Free exercise books": automatic generation of technical exercises
Current CAMIT projects either use manually-written music (be it Mozart or Spice Girls covers), or manually-programmed computer-generated examples (such as in MEWES, and thus MEWES and MEWER).
The current "main" goal of my PhD is to allow the computer to generate its own exercises upon being given a piece (or piece(s)) of music. For example, we could give the computer all the Mozart string quartets, and it would generate rhythmic and pitch exercises for each instrument.
"Understanding the analysis": improved visualization
There are many HCI (human-computer interaction) aspects to consider, and my previous work has included almost none of them. The display in the above screen, for example, is terrible.
I don't know how much HCI fits into my current area of studies, or the research interests of other people here, but I would definitely like to improve this aspect of MEAWS.
"Violin `edutainment'" (I hate that term): Sneaking education into games
The overall goal, bringing together all the above aspects, is educational games. Can we write software to help people learn violin?
The first challenge is to write software that students want to use. The best way to attract interest and retain motivation is to make it a game. Now, there are a wide variety of games -- some people consider merely receiving a score to be a game. Other people might want fancy graphics, an overall plot, non-musical interaction... it varies.
The second challenge is to fit as much educational value. There are plenty of games with very little educational value (for example, the skills learned by playing most 1st person shooter games -- fast reflexes, knowledge of the trade-offs of various imitation or completely imaginary guns, knowledge of the abilities of completely imaginary monsters, etc). There is a small set of games with moderate educational value (such as the old "where in the world is Carmen Sandiego").
Other Interests
I am one of the maintainers of GNU LilyPond.