Find a problem you are passionate about.
That’s the advice MSc grad Greg Burlet has for students thinking about pursuing research in graduate school. For Burlet, his own problem stemmed from a personal interest in music. Computing science student by day, he often collaborated with friend and bandmate Dan Armbruster to write songs outside of class. This process became significantly more complicated when Armbuster relocated to Vancouver.
“We would write guitar riffs and send the audio recordings to each other over email and then try to figure out what the other person was playing on the guitar,” explains Burlet. The process was frustrating for both parties, given how difficult it could be to decipher exactly what the other person was playing and the high likelihood of mistakes. “I thought it would be so much better if I could quickly send sheet music to him describing exactly what I was playing rather than an audio recording.”
And there it was—a problem ripe for solving, and an opportunity for Burlet to combine his two passions of music and artificial intelligence. “It was then that I decided I wanted to pursue graduate school, because I had a burning question in my head that needed to be answered.”
Move over, Siri
Burlet sought this answer through the course of his master’s thesis with the goal of developing software that could listen to guitar music and generate sheet music based on the audio recording.
The first rendition of the software was Robotaba: a website in which users can upload recordings of themselves playing guitar and the program outputs a sheet music translation. “I had an initial vision of the direction I wanted to head in, but I often ran into hurdles that forced me to think critically about the problem and attempt different approaches,” he says. “If the path you took to reach your goal wasn't winding, you probably didn't learn anything—and what's the fun in that?”
Burlet has since improved and enhanced the software into a new, more accurate program called Deep-star Guitar. The software is not unlike voice recognition systems such as Apple’s Siri in that it listens to a single sound source (in this case, a guitar instead of spoken words) and works to determine the notes being played.
“The underlying technology attempts to mimic how the human brain processes and perceives audio by forming a digital model of how neurons fire in the brain,” he explains. “Just as a child learns about his or her environment by seeing examples, this digital model of the human brain learns about the pitch of music notes by being subjected to several examples of guitar recordings.”
“My work allows guitarists to share with the world how they played a piece of music—rather than just the music itself—at the click of a button."
The software was exposed to guitar recordings from several artists including Radiohead, The Beatles, Led Zeppelin, and Metallica. When transcribing the musical notation, the software also determines the most efficient way for a guitarist to play the music to avoid awkward movements or excessive stretching of the hand.
“My work allows guitarists to share with the world how they played a piece of music—rather than just the music itself—at the click of a button,” says Burlet, noting that the technology would be easily marketable as a software application in the commercial sector.
Crossing intellectual borders
When uniting fine art and hard science, one requires a truly versatile perspective. Burlet applauds the Department of Computing Science for its inclusive approach to multidisciplinary research. “The University of Alberta is well-known on a national, and arguably international, scale for its instruction and research in the fields of machine learning and data science,” he says. “It also has an array of professors in computing science who have extremely diverse backgrounds that really foster multidisciplinary research.”
He took several audio courses offered in the Department of Music that provided him with necessary prerequisite knowledge to begin building his software. “I believe that multidisciplinary programs provide several different outlooks on the problems at hand and offer innovative solutions to these problems.”
With a BSc and two master’s degrees behind him, Burlet is now working in the commercial sector as a data scientist for an Edmonton-based startup company called Granify. The work has offered a new and interesting set of problems for solving, but Burlet is up to the challenge. He is also developing a new music transcription app for Android and iOS using his existing software as a foundation.
As for the band? They’re still making music, but he admits it has taken a back seat. “Currently I'm more obsessed with making cool and exciting music software for bands to use.”