Automatic Semantic Characterization Of Drum Sounds
Nowadays, digitized drum kit and percussive samples, loops and beats are ubiquitous and pivotal elements in every sound recording studio, and the overwhelming pace in the dissemination of new sound resources is an evident and growing reality. Withal, the lack of any consistent convention or homogeneous offer (naming conventions, metadata, etc.) shifts a potentially positive profusion into an organizational nightmare, given the necessity of a tedious and cumbersome manual classification for
... ving some benefit of such a huge asset. Consequently, the need of automatic tools for efficient management of this sonic content plethora becomes indisputable. This thesis focuses on the study of the automatic characterization of percussive samples, specifically through the use of semantic descriptors. We start by implementing an automatic taxonomic classification system, in order to find and validate suitable set(s) of features. Commonly, acoustic and psycho-acoustic descriptors are extracted and used for taxonomic and perceptual-based systems. However, semantic approaches to the description of sonic content open a further dimension, which mimics the already established musicians' intuition when describing timbre using adjectives such as bright or clear. In order to address the use of such techniques for percussive sounds, a semantic listening experiment is undertaken, aiming to validate the use of such adjectives and find their perceptual counterparts, traduced in acoustic and psycho-acoustic signal features. Different techniques of regression analysis are used to model the relationship between each of the semantic descriptors and the extracted acoustic and psycho-acoustic descriptors, thus mapping acoustic features into the semantic space.