2010 Aug 20
Hussein Hirjee's and Professor Dan Brown's paper, "Solving Misheard Lyric Search Queries Using a Probabilistic Model of Speech Sounds" has won the best student paper award at the International Society for Music Information Retrieval conference in Utrecht, the Netherlands this year.
It's really easy to mishear song lyrics. For example, if you listen to the beginning of Nirvana's "Smells Like Teen Spirit", you might easily think you'd heard "Don't walk on guns, burn your friends", when the real lyric is, "Load up on guns, bring your friends." Google's no help here: its best match is a Guns N' Roses lyric, from a completely different song. Hussein and Brown's approach instead looks at the constituent sounds of the words and adapts alignment techniques from biosequence analysis to try to find the best match to the phonemes. Their program does significantly better than existing techniques: in a test set, it correctly identifies 85% of songs within the first 10 hits.
About Hussein Hirjee
Hussein is a graduating MMath student who also earned a BCS in 2008 through our undergraduate bioinformatics program. He began his work with Prof. Brown in the summer of 2008. Last year, they published a highly complimented paper at ISMIR on identifying rhyming lyrics in rap music.
About Professor Dan Brown
Professor Brown is having a lot of fun working in a quite new area that is far afield yet surprisingly close to his established expertise in biosequence analysis.
David R. Cheriton School of Computer Science
University of Waterloo
Waterloo, Ontario, Canada N2L 3G1
Tel: 519-888-4567 x33293
Fax: 519-885-1208
Contact | Feedback: newseditor@cs.uwaterloo.ca | David R. Cheriton School of Computer Science | Faculty of Mathematics
Last modified: Friday, 20-Aug-2010 11:59:57 EDT
