Evaluation of Weighted Principal-Component Analysis Matching for Wavetable Synthesis

Wavetable matching of musical instrument tones using principal-component analysis (PCA) takes advantage of spectral correlation information to find the basis spectra. Although PCA matching is efficient, it usually matches the low-amplitude parts of a tone poorly because of its inherent statistical bias. Weighted PCA methods are described, which normalize the tone prior to PCA to fairly weight its different parts. Matching results for a range of instruments show that the PCA of a frame-weighted spectrum with a roughly constant loudness throughout improves on unweighted PCA by an average of about 4% relative spectral error. Listening test results show that frame-weighted PCA gives some perceptual improvements for most of the instruments.