Sinusoids, noise and transients: spectral analysis, feature detection and real-time transformations of audio signals for musical applications


Glover, John C. (2012) Sinusoids, noise and transients: spectral analysis, feature detection and real-time transformations of audio signals for musical applications. PhD thesis, National University of Ireland Maynooth.

[img] Download (6MB)



Add this article to your Mendeley library


Abstract

This thesis examines the possibilities for real-time transformations of musical audio signals using sinusoidal models. Four open-source software libraries were developed with the goal of providing real-time spectral synthesis by analysis tools for composers, musicians, researchers and audio signal processing developers. The first of these, called Simpl, provides a consistent API for interacting with established sinusoidal modelling systems from the literature, and integrates seamlessly with a powerful suite of libraries for scientific computing. Sinusoidal models have been used to transform slowly-varying quasi-harmonic signals with much success, but they have well-documented weaknesses in dealing with transient signal regions. This problem is significant, as in monophonic musical sounds, transient locations often correspond with the attack section of notes and this region plays a large part in our perception of timbre. However in order to improve the synthesis of note attack transients, these regions must first be accurately identified. The first step in this attack transient identification process is generally musical note onset detection. Novel approaches to real-time note onset detection were developed as part of this research and are included in the Modal open-source software library. Modal also includes a set of reference samples that were used to evaluate the performance of the onset detection systems, with the novel methods being shown to perform better than leading solutions from the literature. The onset detection process was then combined with cues taken from the amplitude envelope and the spectral centroid to produce a novel method for segmenting musical tones into attack, sustain and release regions in real-time. The real-time segmentation method was shown to compare favourably with a leading non-real-time technique from the literature, and implementations of both methods are provided in the open-source Note Segmentation library. The Simpl sinusoidal modelling library, Modal onset detection library and Note Segmentation library were then combined to produce a real-time sound manipulation tool called Metamorph. Metamorph is an open source software library for performing high-level sound transformations based on a sinusoids plus noise plus transients model. An overview of the design and implementation of the system is provided, in addition to a collection of examples that demonstrate the functionality of the library and show how it can be extended by creating new sound transformations.

Item Type: Thesis (PhD)
Keywords: Sinusoids; noise; transients; spectral analysis; feature detection; real-time transformations; audio signals; musical applications;
Academic Unit: Faculty of Arts,Celtic Studies and Philosophy > Music
Item ID: 4523
Depositing User: IR eTheses
Date Deposited: 30 Sep 2013 15:58
URI:

    Repository Staff Only(login required)

    View Item Item control page

    Document Downloads

    More statistics for this item...