Digital musicology and computational music analysis have gained momentum in recent years, largely due to the increased creation of symbolic corpora. While covering diverse genres, encodings, formats, and methodologies, most datasets concentrate on melody and/or harmony to infer or predict idiosyncrasies of a certain style, genre, or composer, sometimes also considering musical form. The current project contributes to this trend by analyzing Brazilian Choro, a genre beyond the canon of both classical and popular music research. Choro comprises three meanings: (1) it is a social musical event, (2) it is a musical genre including various subgenres, and (3) it is a manner of performing, a style.
We seek to gain insights into the structural features of harmony and form in this particular genre and relate them to other datasets in order to complement other approaches in corpus-driven music research. Furthermore, we provide visualizations of the formal schema of each piece in this dataset.
We transcribed chord symbols and the formal structure of the 296 songs from the Choro Songbooks (Chediak, Sève, & Souza, 2009, 2011a, 2011b). The transcriptions are based on the notation system by de Clercq & Temperley (2011). They include information about root and bass notes, chord types, added notes, key and meter (global and local) as well as metadata such as genre, year of composition, and composer. We converted all chord roots and bass notes to roman numerals (relative to the local key) and converted the transcriptions into the hierarchical JSON format for statistical analysis and visualization. Subsequently, we applied natural language processing techniques to find patterns on several structural levels (piece, parts, and phrases), and interpreted them as expressing important features of the genre.
Our findings show that frequency vs. rank of chord symbols reveals an underlying Zipf distribution, typically occurring in musical and linguistic corpora. Moreover, prototypical harmonic sequences, such as subdominant-dominant-tonic patterns, make up a large portion of the whole dataset. The formal structure is mainly limited to ternary and rondo forms, and key and modulation patterns vary systematically between subgenres. This, in turn, features can partially be related to diachronic changes within the genre.
The analyses of our dataset support theoretical assertions about patterns of harmony and form in Choro (Almada, 2006). We deduce that these regularities are informative about stylistic characteristics, both for human listeners and algorithms. Our findings can therefore serve as a basis for subsequent music information retrieval applications such as genre or composer prediction within the style in question. They furthermore allow for cross-stylistic comparisons.
Jul 24, 2018 11:10 AM — Jul 24, 2017 11:30 AM
University of Graz