Effect of critical band data reduction on musical instrument sounds

Time-varying spectra of three sets of musical instrument tones were data-reduced using two different methods based on critical bands. One method (used for Tests 1 and 2) preserved the average spectrum and rms amplitude-vs-time envelope for each band. The other method (used for Test 3) utilized critical-band smoothing of the spectral envelope. 20 subjects were tested for their ability to discriminate between original synthesized sounds and data-reduced sounds as a function of a "critical bandwidth multiplier" (CBM). For Test 1 (bassoon, flute, trumpet, harpsichord, harp, marimba, vibraphone, and violin pitched at E b 4), the CBM was varied from 1 to 12, resulting in discriminations between approximately 50% and nearly 100%. However, discrimination at CBM=1 unexpectedly varied from 45 to 85%. For Test 2 (seven piano tones pitched at A0, A1,...,A6), the CBM was varied from 0.25 to 12, resulting in discriminations between 42 and 92%. For Test 3 (bassoon, clarinet, flute, horn, oboe, sax, trumpet, violin pitched at E b 4), the CBM was varied from 1 to 3, resulting in discrimination between 47 and 99%. Our preliminary conclusion is that for CBM=1 the two methods of data reduction cause subtle but discriminable spectral changes for most single musical sounds.