Study of the Timbre Dynamics of an Expressive Performance Using Principal Component Analysis of Spectral Parameters

Hugo B. de Paula

CEFALA - Programa de Pós-graduação em Engenharia Elétrica

Hani C. Yehia

CEFALA - Departamento de Engenharia Eletrônica

Maurício A. Loureiro

CEFALA / CPMC - Escola de Música

CEFALA - Center for Research on Speech, Acoustics, Language and Music

Universidade Federal de Minas Gerais
Av. Pres. Antonio Carlos, 6627
Belo Horizonte, MG, Brasil - 31270-010


Expressive content of performed musical sounds involves the behavior of a wide range of parameters. Timbre dynamics can be responsible for much of the perceived expressiveness of long-note phrases played on wind and bowed instruments. In this paper, Principal Components Analysis (PCA) is used to represent dimensions of spectral dynamics, in order to reveal intentional expressive timbre changes within a gesture realization in a musical performance on the clarinet. Amplitude and frequency time-varying curves of the partials were measured. Spectral envelopes showed a high correlation both in frequency and time domain. PCA has been proved effective in identifying variation patterns in the spectral distribution. The PCA enabled data reduction and transformation making possible to build relations between the principal components and perceptual characteristics of sound. This approach seems to reveal important properties of these parameters that can possibly be characterized as "timbre variation". Results showed that three principal components were enough to represent the "timbre variation" of the samples and the trajectories of these components elucidated some main characteristics of timbre development.