Rhythm and melody in flamenco cantes: challenges for the computational description of flamenco singing
Computational models for rhythmic characterization from musical audio signals mostly rely on measuring patterns (e.g. periodicity) on a set of onset probability functions extracted from music recordings. These probability functions mostly account for fast variations in acoustic descriptors (e.g. energy, pitch or spectral content).
As a consequence, rhythmic description algorithms have strong limitations in vocal music, given the continuous character of the human voice and the variety of pitch ranges and timbre. These limitations are due both to the presence of soft onsets in the voice and to the fact that rhythmic periodicities are commonly less strict in singing compared to other musical instruments. In addition, rhythmic description is strongly linked to melodic characterization when we analyze vocal music, specially a cappella singing.
In this talk I will present our research on computer-assisted singing voice description. In particular I will address the interplay of melody and rhythm in flamenco singing, characterized by the instability of tuning, energy and timbre, often used as an expressive resource. I will illustrate our proposal to characterize a cappella singing and accompanied singing (with guitar) and overview the main technological and ethnomusicological problems that we currently address.