Large-scale corpus analysis of historical electronic music using MIR tools: Informing an ontology of electronic music and cross-validating content-based methods

Dr Nick Collins, Prof Peter Manning, Dr Simone Tarsitani

This project based at Durham University was funded as a 10-month mini-project within the large-scale research project Transforming Musicology, funded under the AHRC’s Digital Transformations scheme, seeking to explore the transformation of musicology through the application of software tools such as those emerging from research in Music Information Retrieval.

Electronic music has developed a rich history over many decades, most intensively since the Second World War, with manifestations in art music and popular music spheres, and much experimental work in-between. The strong heritage of electronic music has been an increasing target of analysts, often featuring certain key works, whether Kontakte, Concret PH or Papa Sangre.

MIR tools offer the possibility to expand this endeavour to a larger database of historical recorded works, tracking audible key trends in compositional and technological endeavour, with an empirical methodology. Ironically, despite the machine-mediated creation of electronic music, automated analysis techniques have not previously been employed to any great degree. Admittedly, although inexhaustible and objective, machine audio analysis has certain limitations compared to the golden standard of human analysis. To provide grounding for this work, we link to an existing project which is supplying human analysis of key works, taking the opportunity to validate machine methods. At the same time, we wish to explore the potential of Linked Data approaches for the release of our findings, including providing a richer context for each work within external databases of historical and social information.

The lead investigator and co-investigator will compile a larger corpus of historical works, aiming for clear coverage of important works in electronic music history, with a balanced approach to experimental art music, and popular music works. The target is around 2000 works, a good compromise between significant corpus size and tractable calculation (if averaging four minutes per piece, this corpus size works out as around 133 hours or five and a half days of audio). We will favour the years 1948-1998, allotting fifty years from the first musique concrète to global electronic dance music before the millennium; this would suggest 400 works per decade as a target, though there will be an inevitable bulge of tracks from the 1970s to 1990s as popular music experimentation really took off and genres proliferated. The selection of works will be informed by the principal and co- investigators’ experience as musicologists of electronic music, between them publishing recent key texts surveying electronic music history, and thus ideally placed as curators for this resource. They will cross reference existing texts, both their own and those of others, for the eligibility of historical pieces. Previous work of the first investigator in particular has laid the groundwork for MIR analysis of electronic music databases of more restricted historical scope (for example, synth pop, early electronic dance music). Yet the scale of the accurately annotated corpus to be created here is novel.

Over ten months, work will proceed following the gradual formation of the corpus, from small- scale experiments of prototype workflow, to large-scale computer runs over the whole database. The meta-data for the corpus, and the output data of experiments, will be encoded to be available as rich linked data online. We aim for deeper musicological insight and theory construction with our work, but further, to provide a core resource for future electronic music scholarship