New dataset captures Brazilian Popular Music’s structure

Researchers created the MPB Corpus, a dataset of 500 pieces of Brazilian Popular Music, known as Musica Popular Brasileira, and detailed annotations of melody, rhythm, harmony, and how melody interacts with harmony.

The corpus offers a resource for computational musicology focused on a genre currently lacking representation in existing datasets. The annotations are symbolic, meaning they describe musical elements rather than providing audio or scores like MIDI files, due to copyright restrictions.

To build the dataset, the team developed analytical models to capture rhythmic and melodic information, alongside tools for visualizing and summarizing musical characteristics. They also performed an initial analysis demonstrating how the corpus can be used to investigate questions about the musical genre. The MPB Corpus is released with an open license, allowing others to use and build upon the data.

Future work could involve applying the dataset to explore the evolution of harmonic and melodic patterns within Brazilian Popular Music.