PERCEPTUAL EVALUATION OF PRINCIPAL-COMPONENT-BASED SYNTHESIS OF MUSICAL TIMBRES

Harmonic-based analysis and resynthesis of musical instrument tones, f or example, using the phase vocoder method, is a valuable technique, b ut its data representation is very large. However, this data set is us ually highly redundant. Principal components analysis (PCA) can be use d to encode such data into a smaller set of orthogonal basis vectors w ith minimal loss of information. Techniques for applying PCA to such d ata are explored, and the aural impact of the method on three tones (c ello, trombone, and clarinet) are studied in two perception experiment s. Results show that nearly identical resyntheses can be produced with a 40-70% data reduction. A preprocessing step called variable-duratio n temporal partitioning (VDTP) is introduced, which also affords a nat ural-sounding method for time expansion and contraction of tones. An e xtension of the PCA technique is also introduced that implements a ''t imbre space,'' or coordinate system for interpolation among a group of musical instruments.