NOTE-WORTHY APP BRINGS AI TO THE MASSES
Like so many inventions, Frettable was born out of necessity.
Greg Burlet dreamt up the artificial intelligence-driven music transcription service after his bandmate moved to B.C.
Still wanting to write music together, the duo started emailing audio files back and forth. The lack of visuals made collaborating difficult, so they tried pointing webcams at their fretboards instead. This too was unsuccessful.
“There was so much lag and the audio quality wasn’t great,” Burlet says.
A University of Alberta undergrad at the time, Burlet had recently taken an introductory class on machine learning and thought: what if a program could recognize the notes played and automatically transcribe them into sheet music or tabulations? It would work like speech recognition, but for music.
His own experience suggested there was a gap in the market, and the idea stuck. The fourth-year computer science student set aside his aspirations of working for video game giant BioWare and decided to pursue not one, but two, master’s degrees in machine learning, which is a subset of artificial intelligence.
While artificial intelligence involves having computers complete tasks that appear to require human intelligence, machine learning enables them to learn and improve from experience the way humans do.
Burlet first received a master’s in music technology from McGill University, then returned to the University of Alberta to complete a master’s in computer science. Though he worked on the intellectual property throughout his post-graduate studies, it wasn’t until he visited Startup Edmonton in fall 2015 that he began building his product.
“Hack days are the perfect environment and ecosystem for creating something from nothing,” Burlet says of the monthly event put on by Startup Edmonton. “You show up early, inject yourself with a whole bunch of caffeine and you’re surrounded by a whole bunch of brilliant people. It’s just a great collaborative brainstorming environment.”
Walking into Edmonton’s iconic Mercer Warehouse he gave himself a goal: to complete the vocal transcription part of his application. By the end of the day, he had hacked together a web interface that could record a person’s voice and indicate which notes were sung.
This was the foundation of the Frettable application, which is now available for both Apple and Android devices and is already in use by several professional musicians.
Intended for a wide audience, the app is sleek, versatile and easy to use. It currently supports vocals, electric guitar, brass and woodwind instruments.
Frettable makes decisions based on the data fed to it — recordings of everything from Radiohead to The Beatles to Burlet’s favourite, Metallica.
“It becomes smarter and smarter the more data you feed it and the longer you train it,” Burlet says.
“That’s the coolest part about [machine learning,] regardless of the domain,” he adds. "It starts off as something that is incredibly naïve about the world and then becomes masterful.”
Burlet chose to build his startup in Edmonton because of his alma mater. Home to the Alberta Machine Intelligence Institute (Amii), the University of Alberta has become the world's academic leader for machine learning.
“There are so many bright individuals that work there and so many professors that are interested in collaborating with industry; it’s kind of a no-brainer to start here,” he says.
By Michelle Ferguson | Photo and Video By Chris Onciul | Additional Footage Provided By Frettable