International Society for Music Information Retrieval


In the final week of last October, the 15th International Society for Music Information Retrieval Conference (ISMIR) took place. For someone who is best labeled as ‘computational musicologist’ (yours truly) this is a relevant event.
Taking the opportunity of having both language and music expertise in the Tunes & Tales project, Folgert Karsdorp and I designed a system that is able to detect cadences (end-of-phrase) in songs using both musical and textual features. Our paper got accepted, and so I visited the conference.

Music Information Retrieval (MIR) is an umbrella term for a quite diverse range of research topics. Of central interest are a number of so called ‘tasks’ such as: genre classification, artist classification, structural segmentation, pattern discovery, beat tracking, music similarity, etcetera. Improving state-of-the-art performance for one of those tasks is generally accepted as a valid contribution to MIR research.


According to the name of the conference, the primary interest is in automatic retrieval of music from (digital) archives or databases. Like Information Retrieval in a more general sense, this can be approached as an engineering task: design a system that given a query returns all relevant documents. There is input and there is output, and there is some processing in between that may remain ‘under the hood’ as long as the system is able to return the desired output for each input query.
Nevertheless, it obviously makes sense to incorporate musical knowledge in these systems. And, vice versa, the process of developing MIR systems might reveal new knowledge about music. That is why musicologists are attending the conference as well. The unavoidable tension between hard-core engineering and musically informed approaches has the potential to raise debate. Talks and papers that address these questions are among the most discussed. A nice example is the paper by Aucouturier and Bigand entitled “Mel Cepstrum & Ann Ova: The Difficult Dialog Between MIR and Music Cognition”, which was presented in 2012.
In that respect, the conference would be a nice case for someone who has a research interest in ‘interdisciplinary’ as such. It would be interesting to send an anthropologist into the ISMIR community to study the tribes, their value systems, their discourses, their open nerves, and their interactions.

Next to the paper about the cadences, I presented the Meertens Tune Collections to the ever data-hungry ISMIR community. These enclose a number of melodic data sets with rich meta data, as presented elsewhere in this magazine.