Working with genres is a very difficult problem. Hundreds of millions of recordings are available on the global music marketplace, so their categorization is necessary to orient curators, editors, promoters, and the audience. However, genres are not well understood since the globalization of music sales on global platforms. Jazz or folk has way too many interpretations and connotations around the globe.

For our feasibility study, we chose folk as a less controversial ‘genre’, but even folk is a too broad category. For the purposes of music promotion, music export, and music documentation, we need taxonomies that work on modern, globalized music platforms. This means that they can be understood for machine learning applications and are resilient in various national contexts from Japan to Lithuania and Vilnius to Rio de Janeiro.

Using ‘folksonomies’, i.e., taxonomies of self-identification by artists and their labels, is not always helpful, and often it can be counter-productive. Genre self-identification by Lithuanian artists in Vilnius may be misunderstood by an algorithm made in the U.S. and trained on global music. Using developed, scientific, and musicological taxonomies may be too slow and expensive for national organizations that want to support all the artists who work in Lithuania.

Global platforms use various quantitative musicological practices to cluster together similar music, i.e., music understood to be jazz in Japan and Lithuania under the ‘jazz’ taxonomy.

The different taxonomic needs of grant agencies (for a folk grant), scientists (for understanding folk), and performing artists (to gain revenues from performances with their folk music) are not conflicting but require attention and education. Sometimes a music project that is successful on world music venues may need to be placed on global streaming platforms under a different taxonomic label. And the artists who may not like the world music label for self-identification may prefer to use a different folksonomy to describe what they do. It is important that our ‘genres’ serve their purpose.

Daniel Antal
Daniel Antal
Data Scientist & Founder of the Digital Music Observatory

My research interests include reproducible social science, finance, and writing R packages.