Discrimination of sustained musical instrument sounds resynthesized with randomly altered harmonic amplitudes

The perceptual salience of random spectrum alteration was investigated for sustained musical instrument sounds. Pitch-synchronous spectral analysis of sounds from eight musical instruments (bassoon, clarinet, flute, horn, oboe, saxophone, trumpet and violin) produced time-varying harmonic amplitude data. The sounds were resynthesized with time-invariant random multipliers applied to the data, yielding average harmonic relative-amplitude errors of 1-50%. In addition, the peak centroids and loudnesses of the altered sounds were equalized to those of the originals. Listeners were asked to discriminate the randomly altered sounds from reference sounds resynthesized from the original data (both resynthesized with flattened harmonic frequencies). For all eight instruments, discrimination was very good for 30-50% errors, moderate for 15-25% errors, and poor for 1-10% errors. Thus, sounds with the same harmonic amplitude-vs.-time envelopes and peak centroid can sound different if the relative-amplitude error is about 15% or more.