Temporal Integration of Auditory Information Is Invariant to Temporal Grouping Cues

Task design. A, For the RT task, subjects indicated their choice (i.e., sequence direction) any time after onset of the sequence. B, For the variable-duration task, subjects indicated their choice after sequence offset. C, For the hybrid task, subjects indicated sequence direction and whether they heard the sequence as one sound or discrete sounds after sequence offset in separate response periods. For all three tasks, subjects were provided feedback regarding frequency direction at the end of each trial (not shown).

Influence of IBI on reports of perceived grouping. Subjects reported whether they perceived the stimulus as one sound or discrete sounds. The graph shows proportion of trials in which each subject chose one sound as a function of IBI. The points indicate each subject’s performance. Each curve represents a logistic function that was fit to each subject’s reports across four sessions. The gray line indicates 50% of trials reported as “one sound”, which was IBI threshold. Colors represent different subjects.

Performance on the RT task. A, The fraction of trials in which subjects reported a sequence of tone bursts was increasing in frequency as a function of signed coherence and IBI. Positive coherence values indicate that the sequence was increasing in frequency; negative indicates decreasing. B, The mean RT (i.e., the time between sequence onset and button press) as a function of signed coherence and IBI on correct trials only (and all 0% coherence trials). C, As in B, but using signal RT (i.e., RT not including cumulative IBI). The solid curves are simultaneous fits of psychometric and chronometric data to a DDM (see Results). The psychometric data (A) only show the fit to the signal-RT data (C). In each panel, the points indicate performance data that was pooled across subjects. Colors represent different IBIs, as indicated in A.

Parameter values from fits of the basic DDM to the RT-task data. Each panel shows best-fitting values of bound height (A) or drift rate (B) plotted as a function of IBI for fits to data from individual subjects (black) or combined across all subjects (red). Dark lines/symbols indicate that the model fits were improved significantly by fitting the given parameter separately for each IBI condition (likelihood-ratio test, p < 0.01, Bonferroni-corrected for three parameters).

LATER model fits to signal RT data. A, Distributions of signal RT from 0%-coherence trials for one subject are plotted on a reciprobit plot: reciprocal RT versus percentage of cumulative frequency on a probit scale (Reddi et al., 2003), separately per IBI. Best-fitting values of the bound height (B) and mean rate-of-rise (C) of the LATER model (see Results and Materials and Methods for details) are plotted as a function of IBI for each subject and coherence (black/gray lines and data points). The data in black indicate that the model fits were improved significantly by fitting the given parameter separately for each IBI condition (likelihood-ratio test, p < 0.01, Bonferroni-corrected for two parameters). Shaded lines/symbols indicate that the model fits were not improved significantly. Red data points/lines represent the median values across all conditions.

Psychophysical kernels. Kernels represent the average of the stimulus sequence (mean subtracted on each trial) presented on all 0%-coherence trials across subjects for increasing (red) and decreasing (blue) choices. A−C, Kernels computed per IBI condition, as indicated, and smoothed using a 21-sample moving mean. D, E, Kernels for the 60 ms IBI condition when subjects were told to emphasize accuracy or speed, respectively. Thick and broken lines are mean and standard error, respectively. Data are aligned relative to onset of the sequence. Asterisks indicate that the increasing and decreasing kernels were significantly different from one another for a given time bin (Mann−Whitney test, p < 0.05^p, using the raw, unsmoothed kernels). Black lines indicate best-fitting simulated kernels (see text for details). Solid vertical lines indicate median RT. Dashed vertical lines indicate the end of the integration time from the best-fitting simulated kernels.

Performance on the variable-duration task. A−C, Psychometric data are plotted as a function of listening duration for different coherences and IBIs, as indicated. Each data point reflects mean performance for all five subjects as a function of coherence and signal time (plotted in 0.2 s bins, up to 1.0 s, but fit using unbinned data). The solid curves are fits from the best-fitting model with two parameters: drift rate and accumulation leak. D, E, Best-fitting values of drift rate (D) and accumulation leak (E) plotted as a function of IBI for fits to data from individual subjects (black) or combined across all subjects (red). Dark lines and symbols indicate that the model fits were improved significantly by fitting the given parameter separately for each IBI condition (likelihood-ratio test, p < 0.01^q, Bonferroni-corrected for two parameters). Shaded lines and symbols indicate that the model fits were not improved significantly.