Automatic Detection of Melodic Patterns in Flamenco Singing by Analyzing Polyphonic Music Recordings

Abstract

In this work an analysis of characteristic melodic pattern in flamenco fandango style is carried out. Contrary to other analysis, where corpora are searched for characteristic melodic patterns, in this work characteristic melodic patterns are defined by flamenco experts and then searched in the corpora. In our case, the corpora were composed of pieces taken from two fandango styles, Valverde fandangos and Huelva capital fandangos. The chosen styles are representative of fandango styles and are also different as for their musical characteristics. The patterns provided by the flamenco experts were specified in MIDI format, but the corpora under study were provided in audio format. Two algorithms had to be designed to accomplish the goal of our research: first, an algorithm extracting audio features from the corpus and outputting a MIDI-like format; second, an algorithm to actually perform the search based on the output provided by the first algorithm. Flamenco experts assessed the results of the searches and drew conclusions.