Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2016 Sep 21;36(38):9908-21.
doi: 10.1523/JNEUROSCI.1421-16.2016.

Slow Temporal Integration Enables Robust Neural Coding and Perception of a Cue to Sound Source Location

Andrew D Brown  1 , Daniel J Tollin  2
Affiliations

Slow Temporal Integration Enables Robust Neural Coding and Perception of a Cue to Sound Source Location

Andrew D Brown et al. J Neurosci. .

Abstract

In mammals, localization of sound sources in azimuth depends on sensitivity to interaural differences in sound timing (ITD) and level (ILD). Paradoxically, while typical ILD-sensitive neurons of the auditory brainstem require millisecond synchrony of excitatory and inhibitory inputs for the encoding of ILDs, human and animal behavioral ILD sensitivity is robust to temporal stimulus degradations (e.g., interaural decorrelation due to reverberation), or, in humans, bilateral clinical device processing. Here we demonstrate that behavioral ILD sensitivity is only modestly degraded with even complete decorrelation of left- and right-ear signals, suggesting the existence of a highly integrative ILD-coding mechanism. Correspondingly, we find that a majority of auditory midbrain neurons in the central nucleus of the inferior colliculus (of chinchilla) effectively encode ILDs despite complete decorrelation of left- and right-ear signals. We show that such responses can be accounted for by relatively long windows of bilateral excitatory-inhibitory interaction, which we explicitly measure using trains of narrowband clicks. Neural and behavioral data are compared with the outputs of a simple model of ILD processing with a single free parameter, the duration of excitatory-inhibitory interaction. Behavioral, neural, and modeling data collectively suggest that ILD sensitivity depends on binaural integration of excitation and inhibition within a ≳3 ms temporal window, significantly longer than observed in lower brainstem neurons. This relatively slow integration potentiates a unique role for the ILD system in spatial hearing that may be of particular importance when informative ITD cues are unavailable.

Significance statement: In mammalian hearing, interaural differences in the timing (ITD) and level (ILD) of impinging sounds carry critical information about source location. However, natural sounds are often decorrelated between the ears by reverberation and background noise, degrading the fidelity of both ITD and ILD cues. Here we demonstrate that behavioral ILD sensitivity (in humans) and neural ILD sensitivity (in single neurons of the chinchilla auditory midbrain) remain robust under stimulus conditions that render ITD cues undetectable. This result can be explained by "slow" temporal integration arising from several-millisecond-long windows of excitatory-inhibitory interaction evident in midbrain, but not brainstem, neurons. Such integrative coding can account for the preservation of ILD sensitivity despite even extreme temporal degradations in ecological acoustic stimuli.

Keywords: binaural hearing; interaural decorrelation; interaural level difference; sound localization; temporal integration.

PubMed Disclaimer

Figures

Figure 1.
Figure 1.
Stimuli of the present investigation. A, Narrowband (one-third octave) noise bursts centered at 4 kHz (psychophysical experiment) or unit CF (physiological experiments) were amplitude-modulated with 100 Hz narrowband noise. B, Narrowband (one-third octave) Gabor clicks centered at 4 kHz (psychophysical experiment) or unit CF (physiological experiments) were concatenated to produce brief click trains of varied ICI. Large ICIs were used in later physiological experiments to yield large contralateral-ipsilateral mismatches, with the temporal mismatch of each click pair specified (see Materials and Methods).
Figure 2.
Figure 2.
Psychophysical ILD sensitivity is robust to interaural decorrelation. A, Subjects performed a dual discrimination and lateralization task, indicating the sidedness and magnitude of the ILD carried by a narrowband target stimulus relative to a diotic reference using a touch-sensitive display. B, Lateralization for correlated (filled symbols) and uncorrelated (open symbols) stimuli across a large range of ILDs. C, Discrimination thresholds for correlated (filled symbols) and uncorrelated (open symbols) stimuli. Error bars indicate ±1 SE.
Figure 3.
Figure 3.
ILD coding in ICC is robust to complete interaural decorrelation of broadband noise. A–D, Mean response rate (±1 SE) across ILD for correlated (filled circles) and uncorrelated (open circles) broadband noise tokens for four neurons of varied CF. Tuning was generally well-described by a four-parameter logistic function (solid line, correlated; dashed line, uncorrelated).
Figure 4.
Figure 4.
ILD tuning and neural discrimination performance in ICC are robust to interaural decorrelation of amplitude-modulated narrowband noise. A, Spike rasters for correlated (top) and uncorrelated (bottom) narrowband noise tokens. The contralateral (putative excitatory) noise token was fixed across conditions, whereas the ipsilateral (putative inhibitory) token was varied. B, Rate-ILD functions for correlated (filled symbols) and uncorrelated (open symbols) narrowband noise, drawn from the rasters in A. Inset, Neural ILD JNDs (heavier lines), derived using Fisher information (lighter lines) for correlated (solid) and uncorrelated (dashed) stimuli.
Figure 5.
Figure 5.
In some ICC neurons, ILD coding persisted for temporally independent click trains. A–D, Rate-ILD functions are shown for 4 neurons presented with temporally identical or independent 250 ms GCTs (see Fig. 1B) with an interclick interval of 10 ms. Legend as in Figure 3.
Figure 6.
Figure 6.
Summary ICC responses across stimulus types demonstrate remarkable robustness to decorrelation. A, Mean logistic rate-ILD functions for neurons tested with correlated (solid line, shading ±1 SE) and uncorrelated (dashed lines) broadband noise. Rates were normalized for each neuron to the maximum of the function for correlated stimuli. Five units lost ILD tuning with decorrelation (numbers in top right; see Results). B, Rate modulation across neurons for correlated (filled circles) and uncorrelated (open circles) broadband noise. A value of 1.0, rarely observed, means that firing was completely inhibited at sufficiently ipsilateral-favoring ILDs. C, Average Fisher information per cell, derived from 1000-repeat bootstrap estimates of total population Fisher information. Error bars indicate bootstrapped 95% CIs for correlated and uncorrelated broadband noise (shading indicates correlated error bars: thin dashed lines indicate uncorrelated error bars). Inset, To calculate population FI, each cell and corresponding rate-ILD function (from Eq. 1; left panel, thin black trace) is assumed to have a mirrored equal (thin gray trace) in the opposite ICC; FI (from Eq. 2) is then taken as the sum of the FI of “left” and “right” cells (middle), from which the neural JND (from Eq. 3) can also be derived (right panel). Across the population, FI per cell is thus the mean of summed “left” and “right” populations divided by the number of cells per population. D, Mean ILD JND per cell, computed from C using Equation 3. E–H, Identical to A–D, but for amplitude-modulated narrowband noise tokens. I–L, Identical to A–D, but for 10 ms ICI GCTs.
Figure 7.
Figure 7.
ICC neurons exhibit several-millisecond duration windows of excitatory-inhibitory interaction. A, An example raster for the 5 s duration 20 ms ICI GCT stimuli used to assess windows of E-I interaction. Despite a +30 dB ipsilateral-favoring ILD, the contralateral stimulus reliably elicits spikes when the ipsilateral (putative inhibitory) stimulus is sufficiently mistimed. B–E, Each panel plots the number of spikes elicited by each of 240 click pairs (smaller points) with given contra (E)–ipsi (I) temporal mismatch. Mean spike count per 1 ms bin of temporal mismatch is given by the larger points, with the size of each mean point scaled by the number contributing observations. Because Gabor clicks had non-negligible duration (dependent on CF), the effective duration of E-I stimulus overlap is indicated by the black bar above each panel. Window width was computed by subtracting this value from the FWHM of a Gaussian fit to mean spike count data (see Materials and Methods, Results).
Figure 8.
Figure 8.
Temporal windows of integration are longer in ICC than in LSO. A, Mean temporal window (±1 SD) for ICC neurons of the present study (black lines) and for 10 LSO neurons from 3 previous studies (asterisk; Joris and Yin, 1995; Park, 1998; Irvine et al., 2001). B, Distributions of window widths in ICC neurons of the present study and previously reported LSO neurons, each computed using Equation 4.
Figure 9.
Figure 9.
Rudimentary model elucidates significance of “slow” temporal integration in ILD processing. A, Inputs to the model, consisting of spike trains generated by the Zilany et al. (2009, 2014) auditory nerve model (see Materials and Methods). In this example, spike trains are shown for 4 ms average ICI GCT stimuli that were binaurally identical (left panel) or independent (two right panels). One channel is assigned a positive (excitatory) value, the other a negative (inhibitory) value. The output is then the sum of E and I within a running temporal window of varied duration (“sliding” bar in each panel), with the result half-wave rectified to eliminate negative spikes (Eq. 5). For example, when stimuli are identical, an inhibition-favoring ILD of 10 dB yields few spikes, even with a short (1 ms) integration window (left). When stimuli are temporally mismatched, however, a large number of spikes leak through the E-I process (middle). A longer (slower) window enables recovery of the inhibition-favoring ILD (right). B, Model outputs for identical and independent GCTs with ICIs of 2, 4, and 8 ms, with window width varied from 1 to 32 ms. Input ILD was varied to enable the generation of simulated rate-ILD curves (normalized to the response rate for the excitatory stimulus alone).
Figure 10.
Figure 10.
Comparison of model predictions and psychophysical data suggests an ILD integration window of ∼3 ms. A, Model responses were used to generate Fisher information-based (thin lines) ILD JNDs (thicker lines) for identical (solid lines) and independent (dashed lines) stimuli. B, A predicted relative ILD JND for each stimulus passed through the model (GCTs of 2, 4, and 8 ms) was generated by normalizing independent to identical model ILD JNDs for each window size. C, A second psychophysical experiment was conducted to measure these values empirically in 5 human subjects. For comparison, model predictions using three different temporal window durations (1, 2, and 4 ms) are indicated (labeled grayscale lines). ILD JNDs increased across ICI, with the observed mean effect falling between model predictions for 2 ms and 4 ms integration windows (left). The same pattern was evident for data from the first psychophysical experiment using narrowband noise stimuli (right).
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Ashida G, Kretzberg J, Tollin DJ. Roles for coincidence detection in coding amplitude-modulated sounds. PLOS Comp Biol. 2016;12:e1004997. doi: 10.1371/journal.pcbi.1004997. - DOI - PMC - PubMed
    1. Astl J, Popelár J, Kvasnák E, Syka J. Comparison of response properties of neurons in the inferior colliculus of guinea pigs under different anesthetics. Int J Audiol. 1996;35:335–345. doi: 10.3109/00206099609071954. - DOI - PubMed
    1. Benevento LA, Coleman PD. Responses of single cells in cat inferior colliculus to binaural click stimuli: combinations of intensity levels, time differences and intensity differences. Brain Res. 1970;17:387–405. doi: 10.1016/0006-8993(70)90248-9. - DOI - PubMed
    1. Boudreau JC, Tsuchitani C. Binaural interaction in the cat superior olive S segment. J Neurophysiol. 1968;31:442–454. - PubMed
    1. Brown AD, Stecker GC. Temporal weighting of interaural time and level differences in high-rate click trains. J Acoust Soc Am. 2010;128:332–341. doi: 10.1121/1.3436540. - DOI - PMC - PubMed

Publication types

LinkOut - more resources