Detecting changes in dynamic and complex acoustic environments

  1. Yves Boubenec  Is a corresponding author
  2. Jennifer Lawlor
  3. Urszula Górska
  4. Shihab Shamma
  5. Bernhard Englitz
  1. Laboratoire des Systèmes Perceptifs, CNRS UMR 8248, France
  2. École normale supérieure, PSL Research University, France
  3. Radboud Universiteit, Netherlands
  4. University of Maryland, United States
11 figures and 4 additional files

Figures

Dynamical change-detection paradigm with auditory textures.

(A) Subjects listened to an acoustic textural stimulus, whose predictability was governed by its marginal frequency distribution (grey curve, left panel). Tones in individual frequency bins were …

https://doi.org/10.7554/eLife.24910.002
Figure 2 with 3 supplements
Detecting a change in statistics improves with size and time of change.

(A) Performance of change detection depended significantly on change time (abscissa) and change size (shades of orange indicate the step size as percent of the original bin probability, see inset). …

https://doi.org/10.7554/eLife.24910.003
Figure 2—figure supplement 1
Change detection improves with base probability.

The prechange marginal probability of a frequency bin significantly influences the performance in the same trial (~10% increase, p=0.005, only 110% condition considered here). Prechange probability …

https://doi.org/10.7554/eLife.24910.004
Figure 2—figure supplement 2
Change detection is not focussed on high probability bins.

Subjects could adopt a strategy to listen to salient, high probability bins. We tested this hypothesis by comparing equal changes in high probability bins, with differential changes in other bins …

https://doi.org/10.7554/eLife.24910.005
Figure 2—figure supplement 3
Change detection improves with stimulus exposure in the previous trial.

(A) Listening duration in the previous trial significantly reduces detectability in the current trial (~15% decrease, p=0.008, Friedman test). A very similar result was obtained in comparison with …

https://doi.org/10.7554/eLife.24910.006
Figure 3 with 1 supplement
Reaction times also reflect estimation of pre- and post-change stimulus properties.

(A) Reaction time distribution sharpens with change size. (B) Median response time significantly reduces by 20% (p<10−4, Kruskal-Wallis) with larger change size (different colors indicate different …

https://doi.org/10.7554/eLife.24910.007
Figure 3—figure supplement 1
Discriminative performance across change sizes.

(A) The probabilities for hits and false alarms were independently computed from their respective reaction time (RT) distributions at each time intervals from 0.2 to 2 s with 0.2 s increments (see …

https://doi.org/10.7554/eLife.24910.008
Detectability of changes depends on spectral properties of the change.

(A) Spectral distance between the changed bin centers ('change distribution', measured in semitones, st) significantly reduces performance (p=0.01, Kruskal-Wallis test). Spectral distance ranged …

https://doi.org/10.7554/eLife.24910.009
Figure 5 with 2 supplements
The CPP potential shows a dependence on both time and size of change, while the central potential remains unaffected.

(A) After stimulus onset, the central potential (Ch. 1, black dot in C) shows a classical N1-P2 progression, followed by a sustained negative potential (labelled NS here). Different shades of red …

https://doi.org/10.7554/eLife.24910.010
Figure 5—figure supplement 1
Change detection performance during the EEG experiment.

(A) The detection rate of subjects in the EEG version of the task was quite comparable to the one in the psychophysics only task (see Figure 2A). (B) The false alarm rate stayed approximately …

https://doi.org/10.7554/eLife.24910.011
Figure 5—figure supplement 2
Same data and analysis as in Figure 5, however, detrended with a classical high-pass filter (Matlab: filtfilt, 0.1 Hz, 15th order, 50 dB attenuation in the stop band).
https://doi.org/10.7554/eLife.24910.012
The CPP potential shows no dependence on whether responses occur early or late after the change.

(A) CPP potentials aligned to response as in Figure 5E2 (for second change-time bin, i.e. around 2.4 s). The solid lines are the early responses (up to median reaction time) and the dashed lines are …

https://doi.org/10.7554/eLife.24910.013
Dual timescale statistical estimation replicates behavioral results.

(A) The dual timescale model consists of two dynamical estimation processes operating with different speeds. If their estimates differ by more than a threshold T, a change in the stimulus is …

https://doi.org/10.7554/eLife.24910.014
A cortical filter-bank model provides an implementation consistent with the behavioral results.

(A) Conceptual structure of the model. The cochleogram (top panel) is passed through modulation filters (scale Ω: 0.54 cycle/oct.; rate ω: 0.72 Hz) for obtaining a cortical representation of the …

https://doi.org/10.7554/eLife.24910.015
Author response image 1
Change detection reaction times and performance during the delayed response EEG experiment as a function of exposure to the first texture Reaction time decreased significantly as a function of change time and trial type both for catch (brown) and change trials (blue, 1 way ANOVA, p-values indicated in the figure).

Reaction times were normalized within each subject before averaging to account for individual overall differences. (A) The accuracy (correct response for either trial type) of catch trials stayed …

https://doi.org/10.7554/eLife.24910.020
Author response image 2
Recreated Figure 5 for the delayed paradigm with a larger number of subjects (n=13), demonstrating that the topography of the potential is unchanged, as are the dependence of slope on change time (which we, however, now interpret as a combination of change time and response time).
https://doi.org/10.7554/eLife.24910.021
Author response image 3
Recreation of Figure 5 for the delayed paradigm with a classical highpass filter, same caption (compare to Author response image 2).
https://doi.org/10.7554/eLife.24910.022

Additional files

Supplementary file 1

Example sounds embedding a change at 3s.

The overall duration of the 4 stimuli is 5 s. Change size is 50%.

https://doi.org/10.7554/eLife.24910.016
Supplementary file 2

Same than Supplementary file 1, with change size 80%.

https://doi.org/10.7554/eLife.24910.017
Supplementary file 3

Same than Supplementary file 1, with change size 110%.

https://doi.org/10.7554/eLife.24910.018
Supplementary file 4

Same than Supplementary file 1, with change size 140%.

https://doi.org/10.7554/eLife.24910.019

Download links