Investigating time-line-based music traditions with field recordings: a case study of Candomblé bell patterns

Abstract

We introduce a series of transdisciplinary corpus studies aimed at investigating cross-cultural trends in time-line-based music traditions. Our analyses concentrate on a compilation of field recordings from the Centre de Recherche en Ethnomusicologie (CREM) sound archive. To demonstrate the value of an interdisciplinary approach combining ethnomusicology and music information research to rhythmic analysis, we propose a case study on the bell patterns used in the musical practices of Candomblé, an Afro-Brazilian religion. After removing vocals from the recordings with a deep learning source separation technique, we further process the instrumental segments using non-negative matrix factorization and select the bell components. Then, we compute a tempo-agnostic rhythmic feature from the bell track and use it to cluster the data. Finally, we use synthesized patterns from the musicological literature about Candomblé as references to propagate labels to the rhythmic clusters in our data. This semi-supervised approach to pattern analysis precludes the need for downbeat and cycle annotations, making it particularly suited for extensive archive investigations. Lastly, by comparing bell patterns in Candomblé and a West African music tradition, we lay the foundation for our future cross-cultural research and observe the potential application of this methodology to other time-line-based music