TITLE: Folk music provides an excellent sandbox and benchmark for machine learning engineering, and reveals its limits!
DATE/TIME: 21st May 2019
LOCATION: School of Informatics. Rm: 4.31/4.33
SPEAKER: Prof. Bob Sturm, KTH, Sweden
Much of my research from the past several years centers around applying machine learning to transcriptions of folk music, specifically Irish, English and Swedish. The resulting models – a union of off-the-shelf machine learning methods and lots of data expressed with a terse representation – are able to generate output good enough that it can almost immediately be plausibly rendered by experts, e.g., see our album and attendant technical report, “Let’s Have Another Gan Ainm” (https://soundcloud.com/oconaillfamilyandfriends). My seminar will recount this work with both engineering and cultural perspectives. An accordion will be played as well.
Bob is an Associate Professor at Stockholm’s Royal Institute of Technology (KTH), where he researches machine listening for music and audio, music modeling and generation, machine learning evaluation, and digital signal processing for sound and music signals. His recent work explores the application of machine learning to model and generate folk music, specifically from Ireland and Sweden. He has degrees in physics, computer music, and electrical engineering.