THESIS
2005
xxiv, 216 leaves : ill. ; 30 cm
Abstract
Musical instrument matching is a classic problem in digital sound synthesis. The goal of instrument matching is to find the best set of parameters to synthesize an acoustic instrument tone. Multiple wavetable synthesis reproduces an instrument tone by adding several enveloped static waveforms stored in wavetables. One successful approach to deciding the basis spectra (that is, the spectra of the wavetables) involves selecting snapshots of the original spectrum. Sophisticated selection strategies such as a genetic algorithm are required to find excellent solutions. An alternative to spectral snapshot selection is to generate the basis spectra, which has received little attention. This thesis describes several artificial basis spectra generation methods based on numerical and statistical...[
Read more ]
Musical instrument matching is a classic problem in digital sound synthesis. The goal of instrument matching is to find the best set of parameters to synthesize an acoustic instrument tone. Multiple wavetable synthesis reproduces an instrument tone by adding several enveloped static waveforms stored in wavetables. One successful approach to deciding the basis spectra (that is, the spectra of the wavetables) involves selecting snapshots of the original spectrum. Sophisticated selection strategies such as a genetic algorithm are required to find excellent solutions. An alternative to spectral snapshot selection is to generate the basis spectra, which has received little attention. This thesis describes several artificial basis spectra generation methods based on numerical and statistical techniques. The matching results for a range of instruments show that a local search method slightly outperforms genetic algorithm selection, and is more efficient and easier to program; and that a weighted principal components analysis and other iterative methods, which are significantly faster, can find matches comparable to near-optimal non-iterative matches. Our results also suggest some important characteristics of the wavetable search space. The new methods constitute a practical set that meets different specifications for synthesizers, and can be extended to multi-note wavetable matching.
Post a Comment