Structure Preserving Transformations in the Comparison of Complex, Steady-State Sounds.

摘要

A process-oriented feature selection model was proposed to characterize listeners' comparisons of complex sounds. Specifically, the model assumes that the listener performs a structural analysis on the low-resolution spectra of the stimuli to be compared and then extracts a feature representation through a structure-preserving transformation resembling a principal-components analysis. This feature representation is subsequently employed to make similarity judgments between stimuli. Predictions of the model for a timbre-comparison task were examined using a set of sixteen complex sounds that varied in amplitude-spectral shape. The subjective feature representation obtained from the ALSCAL nonmetric scaling program was generally consistent with the theoretical feature representation produced by the optimal structure-preserving transformation applied to the loudness-weighted spectra. The two comparison features as well as the relative importance of the two dimensions were successfully predicted by the model. Practical implications for the subjective evaluation of complex signals are discussed and refinements to the transformations in the model are suggested for further research. (Author)