MPB Corpus: 500 Brazilian songs encoded for AI musicology research

The MPB Corpus, introduced in a new arXiv paper by Carlos de L. Almada, Hugo T. de Carvalho, and Felipe D. Martins, delivers 500 Brazilian popular music (MPB) pieces encoded across four core musical parameters: melodic contour, melodic rhythm, harmony, and the relationship between melody and harmony. This makes it the most extensive and detailed dataset yet created for computational musicology focused on Brazilian music. The encoding process relies on purpose-built analytical models that capture rhythmic and melodic information with high precision, paired with tailored visualizations and metrics that summarize each musical parameter.

The paper (arXiv:2608.13842) spans 22 pages and includes 13 figures, offering a qualitative exploratory analysis that demonstrates how the corpus can be used to both formulate and systematically address musicological questions about MPB. As an open resource, the dataset is poised to support research in sound computing, digital libraries, information retrieval, and audio processing. For AI researchers, the MPB Corpus provides a structured, high-quality ground truth for training models on melody-harmony relationships, rhythm analysis, and genre-specific musical structures—potentially advancing generative music systems that understand Brazilian musical idioms.