Perceptual evaluations of synthesized musical instrument tones

An analysis-based synthesis technique for the computer generation of musical instrument tones was perpetually evaluated in terms of the discriminability of 16 original and resynthesized tones taken from a wide class of orchestral instruments having quasiharmonic series. The analysis technique used was the heterodyne filter—which produced a set of intermediate data for additive synthesis, consisting of time-varying amplitude and frequency functions for the set of partials of each tone. Three successive levels of data reduction were applied to this intermediate data, producing types of simplified signals that were also compared with the original and resynthesized tones. The results of this study, in which all combinations of signals were compared, demonstrated the perceptual closeness of the original and directly resynthesized tones. An orderly relationship was found between the form of data reduction incurred by the signals and their relative discriminability, measured by a modified AAAB discrimination procedure. Direct judgments of the relative sizes of the differences between the tones agreed with these results; multidimensional scaling of the latter data provided a visual display of the relationships among the different forms of tones and pointed out the importance of certain small details existing in the attack segments of tones. Of the three forms of simplification attempted with the tones, the most successful was a line-segment approximation to time-varying amplitude and frequency functions for the partials. The pronounced success of this modification strongly suggests that many of the microfluctuations usually found in the analyzed physical attributes of music instrument tones have little perceptual significance.