Dataset Forró em Vinil

In the work described in both the paper Measuring disruption in song similarity networks and the doctoral thesis entitled Caracterizando e Quantificando Inovação em Redes de Similaridade Musical (only available in portuguese), Falcão et al. developed a study of disruption measurement over a collection of a Brazilian music tradition called Forró.

This dataset contains all the audio information used during analysis. 25,605 audio files were used to build a song similarity network, from which disruption information was derived.

list_of_songs.txt, mfccs.csv and latent_representations.csv contain an indexed list of all the audio files analysed, their MFCC-based feature vectors and a simplified latent representation of the MFCCs generated by an autoencoder developed during the study, respectively. Data from all these files can be mapped by using the indexes (i.e., the i-eth line in mfccs.csv refers to the feature vector for the i-eth song informed by list_of_songs.txt)

similarity network - unweighted.gexf contains the similarity network built according to the similarities calculated using the feature vectors. The network is available in a GEXF (Graph Exchange XML Format) format, and can be visualized with softwares like Gephi.