This talk introduces the WITCHCRAFT project, a 4-year project to design and build a usable melody search engine for folksongs. It will start on May 1st 2006, so right after this seminar. I therefore cannot present much results yet. Instead I will discuss the motivation of the project, the materials we are going to use, and some of the issues we expect to encounter.
The motivation for the WITCHCRAFT project comes to a significant extent from the cultural heritage partner, the Meertens Instituut for research and documentation of Dutch language and culture. It maintains a database of Dutch folksongs (the Liederenbank) that contains a large amount of metadata. But there is no content-based melody search, and this is what the project will remedy. Specifically it will provide content-based access to a large collection of melodies known as Onder de Groene Linde (ca. 7000 songs). Researchers need content based searching to help them identify melodies and to trace to changes a melody underwent over time and in space. But also the general public is genuinely interested in this repertoire, so the usability of the system is an important issue. The variability of melody, caused by oral transmission, is probably the most complex problem to deal with. To be able to do so we need to integrate quite some music cognition knowledge into the system. I will conclude my talk by performing a small experiment that illustrates the role of human memory in melody recollection.