Ashok K. Krishnamurthy
Lawrence L. Feth
Dept. of Elec. Eng. & Div. of Speech and Hear. Sci., Ohio State Univ., Columbus, OH 43210
The intensity/envelope weighted average of instantaneous frequency (IWAIF/EWAIF) has been used to model listener performance in complex signal discrimination [Feth, J. Acoust. Soc. Am. Suppl. 1 88, S48 (1990)]. Previous work with this model has used the entire signal duration for the IWAIF computation. However there exist pairs of signals such as the STEP and GLIDE signals used by Madden and Feth [Feth et al., J. Acoust. Soc. Am. Suppl. 1 86, S122 (1989); Madden and Feth, J. Speech Hear. Res. 35, 436--442 (1992)] that have the same long-term IWAIF values, but are nonetheless readily discriminable by listeners. The addition of a short-term running IWAIF computation extends the application of the IWAIF model to such dynamic signals. Also described is an ideal detector for the short-term IWAIF model that assumes that independent, white, Gaussian internal noise is added after the IWAIF computation. The noise variance is derived from frequency DL (FDL) data. It is shown that this detector is able to predict listener performance in the STEP versus GLIDE discrimination task. Applications of the short-term IWAIF model in deriving frequency modulation DL (FMDL) from the FDL are also presented. [Work supported by AFOSR.]