Sonological models for timbre characterization

In the research on timbre, two important variables have to be assigned from the onset: the instruments used to analyze and to model the physical sound, and the techniques employed to provide an efficient and manageable representation of the data. The experimental methodology which results from these choices defines a {em sonological model}; several different psychoacoustical and analytical tools have been employed to this aim in the past.
In this paper we will present a series of experiments conducted at the CSC-University of Padova, which attempted to define an experimental framework for the development of algorithmically-defined timbre spaces. Fundamental to our line of research has been the use of analysis methods borrowed from the speech processing community and of data-representation techniques such as neural networks and statistical tools.
The results show several analogies to the classical timbre spaces defined in the literature; this has proved very important in order to explore the qualities of musical timbre in a purely analytical way which does not rely on subjective listeners' ratings.