Computer Sound Transformations Guided by Perceptually Motivated Features

Abstract

In a time where technology is part of our everyday life, sound and music is coming to us, more and more, in digital format. But this vast amount of digital audio available demands a deeper understanding of audio signals, in particular how algorithms are formulated so that the information of the audio data can be extracted automatically. A big challenge when developing an audio information retrieval system is the identification of appropriate content-based features to represent the audio signal. The most common approach to represent the audio in such system is using audio descriptors, which measures properties of audio signal content and wrap audio features to sets of values. This descriptors can be divided in three levels: low, medium and high. This levels will be detailed later. This dissertation starts by a study of the state-of-the-art on Music Information Retrieval and systems used to get relevant information from audio files. Within this scope, I surveyed two models found in the literature to describe mathematically the warmth of musical audio. After revising the latter sound warmth metric of the studies I aim to provide a better mathematical model to the descriptor finding the constant that better define a linear correlation between user judgments of sound warmth and the two low-level sound descriptors proposed in the studies found. In sum, this dissertation proposes a (one-knob) audio effect which allows users to transform the warmth of a sound in real-time.