Schematic of the noise-band vocoder used to process the lead–lag stimuli which were spatialized with HRTFs. Interaural correlation of the carrier noise was varied (cf. Sec. II D; LP, low-pass filter; NBN, narrow-band noise).
Top: Localization results as a function of the delay between lead and lag for a low-pass noise burst (left column), a wide-band noise burst (middle), and the word “shape” (right). As plotted, the lead was played from +30° and the lag from −30° azimuth, but in the experiment lead and lag locations were randomized. In one session subjects were instructed to point to the rightmost image if they heard two or more sounds, and in another session to the leftmost image. When the lead was on the left and subjects were instructed to point to the leftmost image, they effectively pointed to the lead. These data were sign-inverted and combined with the data for pointing to the right when the lead was on the right and plotted as the lead image. Likewise, the lag image was formed from data when the instruction was to point to the leftmost image and the lag was on the left. 120 responses from six subjects are plotted per delay. Medians and quartiles of the pooled results are given for the lead and lag image. Bottom: Subjects were instructed to press a different button according to whether they perceived one or more than one image, i.e. whether the lag was audible. The lower subplots show the percentage of responses for hearing two or more images. The solid line indicates data from pointing to the lead while the dashed-dotted line indicates those from the lag. Only one line is visible because of their coincidence.
Lateralization results for vocoded lead–lag stimuli; a low-pass noise burst (left column) and a wide-band noise burst (right). Results are plotted similarly to Fig. 2 as medians and quartiles of responses parsed for pointing to the lead or lag image. Note that ±1 depicts maximum lateralization at the ears and not the lead or lag position. The top row presents results for uncorrelated carriers and the bottom row for correlated carriers. The lower subplots show the percentage of responses for hearing two or more images with the solid line indicating data from pointing to the lead and the dashed-dotted line those from the lag. Note that both lines are almost coincident.
Same as Fig. 3, but for the word “shape” and five different values of the carrier correlation given in the inset. Irrespective of carrier correlation, a lag image was reported at almost all delays, indicating that the PE failed.
Same as Fig. 4, but for the words “wide” (left) and “teak” (right) for uncorrelated (top) and correlated carriers (bottom). The lag appeared clearly separated from 2 to 6 ms delay for “wide” and from 6 to 12 ms for “teak.”
Lateralization dominance results for the word “shape” for uncorrelated carriers when subjects were instructed to point to the most dominant and the weakest image. Results for the lead and lag image were replotted from Fig. 4 in gray without error bars. The lead and dominant images coincided while the weakest image largely matched the lag image, indicating reduced perceptual salience of the lag.
Proportion of responses on the side of the lag for all stimuli and conditions as a function of delay time. The thick line refers to results with unprocessed stimuli, the thin line to vocoded conditions with uncorrelated carriers, the dashed line with correlated carriers and the dotted lines with carriers with correlations between zero and one. For unprocessed stimuli almost no responses fell on the side of the lag at short delays, indicating strong localization dominance of the lead. After vocoding, many responses occurred on the lag side at similar delays, indicating strongly reduced lead dominance.
ETs from unprocessed (horizontal lines without markers) and vocoded (with markers) conditions computed from either localization data (L-ET, solid lines) or from responses for hearing more than one image (P-ETs, dashed lines). For vocoded conditions ETs are given as a function of carrier correlation.
Proportion of participants for which statistical testing indicated that responses cluster significantly into two images rather than one as a function of delay time. Cluster analysis was used to fit one or two Normal distributions to the pooled results for the lead and lag image of each subject. Given is the proportion of subjects for which the fit with two clusters resulted in a significantly higher loglikelihood than that with one cluster (p < 0.01).
Article metrics loading...
Full text loading...