Discrimination of formant frequency, intensity, and bandwidth in natural speech

Abstract

Psychoacoustical discrimination limits described the ability of the human observer to perceive changes in an acoustic stimulus; sounds differing by less than a “Just Noticeable Difference” (JND) are similarly heard. Flanagan and his associates [J. Acoust. Soc. Am. 27, 613 (1955); 27, 1223 (1955); 30, 435 (1958)] studied steady‐state synthetic speech sounds and measured various spectrally defined JND's. Klatt [J. Acoust. Soc. Am. 53, 8 (1973)] reexamined the JND for F0 using synthetic speech sounds that modeled the time varying features of F0 in natural speech and found it about an order of magnitude larger than reported earlier. We consider perceptually equivalent speech sounds, when all the JND's described by Flanagan are taken into account. To simplify the task of finding the JND's appropriate for natural speech we assume they all have a similar dependence upon frequency and that they are similar multiples of the steady‐state JND's; the detailed dependence upon spectral frequency was taken from studies of periodicity pitch [Goldstein, J. Acoust. Soc. Am. 54, 1496 (1973)]. Speech that was anlyzed and resynthesized using a Residual Adaptive Predictive Coder was studied psychoacoustically. The quantization precision of the coder coefficients was chosen so that the resynthesized speech would not differ from the original by more than a given JND. A distance measure based upon formant JND's was formulated, and a quantization table with which to quantize the coder coefficients was constructed. Synthetic speech samples were prepared assuming various different increases of the JND's over their steady‐state values and listeners scored the speech quality. Experiments indicate that the JND's are about five times greater than the values measured earlier by Flanagan.