ISMIR paper

Today I finished rewriting  a paper for ISMIR 2009. ISMIR is an annual conference on Music Information Retrieval. This year, it will be held in Kobe, Japan, during the last week of October.

The paper is entitled Musical Models for Folk-Song Melody Alignment. I’m quite satisfied with this title; it reflects a very important basic principle: in order to use methods from Computer Science for music processing, one needs to develop models of musical knowledge. This sounds pretty straightforward, but it appears to be very difficult to evaluate the musical meaning of the results of algorithms that are applied to musical data. In most studies the aim is merely to reproduce a certain fixed “ground-truth”. It can be quite an achievement if that succeeds, but it is far more interesting to see what we can learn about ground-truth data itself using computational methods. Of course, it then is not “ground-truth” anymore, but data to be studied.

The method we use in this paper (sequence alignment) has been developed within Computer Science, but not specifically for music. In order to use it to align folk-song melodies, we incorporated some (simple) musical models in the algorithm’s scoring functions. Those functions are based on musicological knowledge about similarity of folk-songs. In this paper we still evaluate the algorithm using a ground-truth, but the next step will be to evaluate the ground-truth using the algorithm. I really look forward to the results of that.