Communications of the ACM
Lennon or McCartney? Machine Learning Tries to Crack Disputed Beatles Authorship
Researchers at Harvard University and Canada's Dalhousie University used machine learning to ascertain the authorship of disputed Beatles songs.
Credit: Getty
Researchers at Harvard University and Canada's Dalhousie University used machine learning to ascertain the authorship of disputed Beatles songs.
Over a three-year period, the team created an algorithm and applied it to an array of musical motifs taken from 70 songs co-authored by Paul McCartney and John Lennon.
The algorithm was able to distinguish tracks known to be written by one of the two musicians with 76% accuracy.
When applied to eight disputed songs or song segments, the algorithm found a number of the songs yielded a more than 90% match against Lennon's other works; one song was an 81% match for Lennon, with its bridge returning a 57% match for McCartney.
Harvard's Mark Glickman hopes the algorithm's statistical model finds use as "a blueprint for those wanting to follow changes in music over time."
From Financial Times
View Full Article - May Require Paid Subscription
Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA
No entries found