|
Title
Automatic Sound Synthesizer Programming: Techniques and Applications
Local download of dphil thesis: PDF of thesis
Abstract
The aim of this thesis is to investigate techniques for, and applications of automatic sound
synthesizer programming. An automatic sound synthesizer programmer is a system which
removes the requirement to explicitly specify parameter settings for a sound synthesis
algorithm from the user. Two forms of these systems are discussed in this thesis: tone
matching programmers and synthesis space explorers. A tone matching programmer takes
at its input a sound synthesis algorithm and a desired target sound. At its output it pro-
duces a conguration for the sound synthesis algorithm which causes it to emit a similar
sound to the target. The techniques for achieving this that are investigated are genetic
algorithms, neural networks, hill climbers and data driven approaches. A synthesis space
explorer provides a user with a representation of the space of possible sounds that a syn-
thesizer can produce and allows them to interactively explore this space. The applications
of automatic sound synthesizer programming that are investigated include studio tools,
an autonomous musical agent and a self-reprogramming drum machine. The research em-
ploys several methodologies: the development of novel software frameworks and tools, the
examination of existing software at the source code and performance levels and user trials
of the tools and software. The main contributions made are: a method for visualisation of
sound synthesis space and low dimensional control of sound synthesizers; a general purpose
framework for the deployment and testing of sound synthesis and optimisation algorithms
in the SuperCollider language sclang; a comparison of a variety of optimisation techniques
for sound synthesizer programming; an analysis of sound synthesizer error surfaces; a gen-
eral purpose sound synthesizer programmer compatible with industry standard tools; an
automatic improviser which passes a loose equivalent of the Turing test for Jazz musicians,
i.e. being half of a man-machine duet which was rated as one of the best sessions of 2009
on the BBC's 'Jazz on 3' programme.
top
|
|