DISCRIMINATION OF RANDOM, TIME-VARYING SPECTRA WITH STATISTICAL CONSTRAINTS

In the present study, a sample-discrimination procedure is used to exa mine how various statistical constraints on random, time-varying spect ra might aid listeners in discriminating among these signals. Time-var ying spectra were constructed by adding five sequences of tones, five tones in each sequence, centered at the octave frequencies from 250 to 4000 Hz. The levels, frequencies, and durations of the tones varied r andomly from one presentation to the next. The signal to be detected w as an increment in the level of tones comprising the sequence centered at 1000 Hz (the target sequence). Three main conditions were examined in which the parameter values of tones varied independently of one an other (equal-variance condition), covaried across tone sequences (cros s-channel covariance condition), or covaried among themselves for each tone (cross-parameter covariance condition). For the latter two condi tions, two special cases were examined in which the target sequence va ried independently of nontarget sequences or varied according to the s ame statistical rules. For five of six listeners, comparisons of perfo rmance to that of an ideal, within-channel observer gave clear evidenc e of an ability to take advantage of cross-channel statistical constra ints, but only when the target sequence varied independently of its co ntext. Results offer no evidence for listeners' ability to take advant age of cross-parameter constraints. Trial-by-trial analyses in the equ al-variance condition identify factors other than the appropriate weig hting of information as primarily responsible for suboptimal performan ce, though individual differences complicate this interpretation. A fa ilure of simple detection-theoretic models to account for these result s is discussed in terms of possible perceptual rules governing sound s ource identification of real-world objects.