Tools for detection and classification of piano drum patterns from Candombe recordings
This work is part of an interdisciplinary collaboration that pursues the development of automatic tools for computer–aided analysis and transcription of Candombe rhythm patterns from audio recordings.
Martín Rocamora, Luis Jure and Luiz W. P. Biscainho
December 4, 2014
Berlin, Germany
9th Conference on Interdisciplinary Musicology — CIM14
Abstract
The African-rooted Candombe drumming is one of the most characteristic traits of Uruguayan culture. Yet its complex rhythm can sound difficult to decode for unfamiliar listeners. Supported by over two decades of systematic study of Candombe from a musicological perspective, this work is part of an interdisciplinary collaboration that pursues the development of automatic tools for computer-aided analysis and transcription of Candombe from audio recordings. Having previously devised a supervised scheme for rhythmic pattern tracking intended for finding the underlying metrical structure of recorded music, in the present work the authors propose a technique for detailed analysis of the rhythmic patterns of each type of drum as found in recorded Candombe performances. The platform for investigation was an audio database composed of more than 20 recordings involving about a dozen players (totaling over an hour of audio), whose metrical structures have been identified and annotated by an expert. For a given recording, the first proposed tool consists in constructing a map of bar-length patterns that enables the inspection of their evolution over time; a second tool clusters those patterns, and maps the result onto a low dimensional space for visualization. After a set of controlled experiments that illustrate and justify the fundamentals of the described strategies, they are applied to the analysis of real piano drum performances as a case study. Among other findings, the tests demonstrated that the obtained clusters match characteristic patterns of the instrument; moreover, when one compares recordings by different performers the patterns tend to be grouped by artist, thus disclosing differences and similarities among their personal styles. The ultimate goal of this investigation is to provide an interactive software tool for the analysis of Candombe recordings oriented towards the study of the underlying structures and rules governing performance styles and improvisation.