Discrimination of dynamic frequency sounds and a multichannel detector model

Abstract

A virtual frequency (VF) glide signal refers to the frequency transition perceived in an amplitude-modulated two-tone complex. For example, when the amplitude of one tone increases with duration, while that of the other decreases, the listener hears a rising pitch. This virtual glide is easily distinguished from a linear frequency-modulated (FM) glide because of envelope fluctuations in the virtual glide. They can be rendered more difficult to discriminate by extracting the envelope from the VF glide and imposing it on the FM glide. Discrimination of common envelope VF and FM glides was studied through an adaptive 2Q, 2-AFC task. VF and FM glides (250 ms) were centered on 500-, 1000-, 2000-, and 4000-kHz. Frequency sweeps began at approximately 1/7 octave and were adjusted using the 2 up, 1 Down rule (Levitt, 1971). Center frequency roved over a range of approximately 1/2 octave for each interval. We show that a multichannel detector model (Durlach, 1986) operating on intensity weighted average of instantaneous frequency (IWAIF) values computed by a short-term, multichannel IWAIF model is able to predict listener performance in the common-envelope VF versus FM glide discrimination task. The noise variance required by the model is derived from measured FMDL data.