フォルマント状スペクトルをもつ合成音に対する音色弁別について (<音声特集号>)

抄録

After finishing the course of our quality study of speech sounds, we propose here the quality problem of synthesized sounds having formant-like spectra. The immediate aim of our present experiment, is devided into two: one is to determine the discriminatory bounds in quality response in regards to the changes in formant position and formant band width; the other is to illuminate the mechanism of the formation of so-called mental map by which the identification of several sounds which differ in both position and width of formant is established. In order to synthesize a formant-like complex sounds, damped sinusoids with repetition rate of 200 waves per second are made in use by introducing a narrow pulse-train through a single resonant circuit. The following points can be safely concluded as basic properties in timbre judgement. (1) The ratio of the insensitivity σ__<f_0 > to the half-power band width B regarding the formant frequency, defined by the standard deviation of distribution of equal judgement in AB listening test, is approximately constant regardless the resonant frequency f_0. (2) The ratio of the insensitivity σ_B to the formant band width B is generally constant. (3) When considering the confusion matrix rather from the discriminatory view-point, than from the view-point of identification, the insensitivities can be obtained from the confusion data in two directions, i. e. , in outgoing and incoming one. These are far greater than that observed in the discrimination test of pair-comparison type. A member of sound group does not necessarily have same values of insensitivities in two directions. (4) Definiteness of qualities of formant-like sounds, which are correctly identified, cannot be reflected directly on the magnitude of insensitivity defined from the discriminatory view-point.