A framework for testing and comparing binaural models

A comparative study of eight human auditory models of monaural processing

A number of auditory models have been developed using diverging approaches, either physiological or perceptual, but they share comparable stages of signal processing, as they are inspired by the same constitutive parts of the auditory system. We compare eight monaural models that are openly accessible in the Auditory Modelling Toolbox. We discuss the considerations required to make the model outputs comparable to each other, as well as the results for the following model processing stages or their equivalents: Outer and middle ear, cochlear filter bank, inner hair cell, auditory nerve synapse, cochlear nucleus, and inferior colliculus. The discussion includes a list of recommendations for future applications of auditory models.

Testing models at the neural level reveals how the brain computes subjective value

In recent years, models have played an increasingly important role for understanding the brain in cognitive, behavioral, and systems neuroscience. Decision neuroscience in particular has benefitted greatly from the application of economic models of choice preferences to neural data. However, an often-overlooked aspect is that many models of preferences have a generic problem—they make extremely similar behavioral predictions. Here, we demonstrate that to understand the mechanisms of valuation in the brain, it is useful to compare models of choice preferences not only at the behavioral but also at the neural level.

Decisions are based on the subjective values of choice options. However, subjective value is a theoretical construct and not directly observable. Strikingly, distinct theoretical models competing to explain how subjective values are assigned to choice options often make very similar behavioral predictions, which poses a major difficulty for establishing a mechanistic, biologically plausible explanation of decision-making based on behavior alone. Here, we demonstrate that model comparison at the neural level provides insights into model implementation during subjective value computation even though the distinct models parametrically identify common brain regions as computing subjective value. We show that frontal cortical regions implement a model based on the statistical distributions of available rewards, whereas intraparietal cortex and striatum compute subjective value signals according to a model based on distortions in the representations of probabilities. Thus, better mechanistic understanding of how cognitive processes are implemented arises from model comparisons at the neural level, over and above the traditional approach of comparing models at the behavioral level alone.

Robustness of neuronal tuning to binaural sound localization cues against age-related loss of inhibitory synaptic inputs

Sound localization relies on minute differences in the timing and intensity of sound arriving at both ears. Neurons of the lateral superior olive (LSO) in the brainstem process these interaural disparities by precisely detecting excitatory and inhibitory synaptic inputs. Aging generally induces selective loss of inhibitory synaptic transmission along the entire auditory pathways, including the reduction of inhibitory afferents to LSO. Electrophysiological recordings in animals, however, reported only minor functional changes in aged LSO. The perplexing discrepancy between anatomical and physiological observations suggests a role for activity-dependent plasticity that would help neurons retain their binaural tuning function despite loss of inhibitory inputs. To explore this hypothesis, we use a computational model of LSO to investigate mechanisms underlying the observed functional robustness against age-related loss of inhibitory inputs. The LSO model is an integrate-and-fire type enhanced with a small amount of low-voltage activated potassium conductance and driven with (in)homogeneous Poissonian inputs. Without synaptic input loss, model spike rates varied smoothly with interaural time and level differences, replicating empirical tuning properties of LSO. By reducing the number of inhibitory afferents to mimic age-related loss of inhibition, overall spike rates increased, which negatively impacted binaural tuning performance, measured as modulation depth and neuronal discriminability. To simulate a recovery process compensating for the loss of inhibitory fibers, the strength of remaining inhibitory inputs was increased. By this modification, effects of inhibition loss on binaural tuning were considerably weakened, leading to an improvement of functional performance. These neuron-level observations were further confirmed by population modeling, in which binaural tuning properties of multiple LSO neurons were varied according to empirical measurements. These results demonstrate the plausibility that homeostatic plasticity could effectively counteract known age-dependent loss of inhibitory fibers in LSO and suggest that behavioral degradation of sound localization might originate from changes occurring more centrally.

Neural rate difference model can account for lateralization of high-frequency stimuli

Lateralization of complex high-frequency sounds is conveyed by interaural level differences (ILDs) and interaural time differences (ITDs) in the envelope. In this work, the authors constructed an auditory model and simulate data from three previous behavioral studies obtained with, in total, over 1000 different amplitude-modulated stimuli. The authors combine a well-established auditory periphery model with a functional count-comparison model for binaural excitatory-inhibitory (EI) interaction. After parameter optimization of the EI-model stage, the hemispheric rate-difference between pairs of EI-model neurons relates linearly with the extent of laterality in human listeners. If a certain ILD and a certain envelope ITD each cause a similar extent of laterality, they also produce a similar rate difference in the same model neurons. After parameter optimization, the model accounts for 95.7% of the variance in the largest dataset, in which amplitude modulation depth, rate of modulation, modulation exponent, ILD, and envelope ITD were varied. The model also accounts for 83% of the variances in each of the other two datasets using the same EI model parameters.

Sound localization in barn owls studied with manipulated head-related transfer functions: beyond broadband interaural time and level differences

Interaural time and level differences are important cues for sound localization. We wondered whether the broadband information contained in these two cues could fully explain the behavior of barn owls and responses of midbrain neurons in these birds. To tackle this problem, we developed a novel approach based on head-related transfer functions. These filters contain the complete information present at the eardrum. We selected positions in space characterized by equal broadband interaural time and level differences. Stimulation from such positions provides reduced information to the owl. We show that barn owls are able to discriminate between such positions. In many cases, but not all, the owls may have used spectral components of interaural level differences that exceeded the known behavioral resolution and variability for discrimination. Alternatively, the birds may have used template matching. Likewise, neurons in the optic tectum of the barn owl, a nucleus involved in sensorimotor integration, contained more information than is available in the broadband interaural time and level differences. Thus, these data show that more information is available and used by barn owls for sound localization than carried by broadband interaural time and level differences.

Population rate-coding predicts correctly that human sound localization depends on sound intensity

Human sound localization is an important computation performed by the brain. Models of sound localization commonly assume that sound lateralization from interaural time differences is level invariant. Here we observe that two prevalent theories of sound localization make opposing predictions. The labelled-line model encodes location through tuned representations of spatial location and predicts that perceived direction is level invariant. In contrast, the hemispheric-difference model encodes location through spike-rate and predicts that perceived direction becomes medially biased at low sound levels. Here, behavioral experiments find that softer sounds are perceived closer to midline than louder sounds, favoring rate-coding models of human sound localization. Analogously, visual depth perception, which is based on interocular disparity, depends on the contrast of the target. The similar results in hearing and vision suggest that the brain may use a canonical computation of location: encoding perceived location through population spike rate relative to baseline.

Being able to localize sounds helps us make sense of the world around us. The brain works out sound direction by comparing the times of when sound reaches the left versus the right ear. This cue is known as interaural time difference, or ITD for short. But how exactly the brain decodes this information is still unknown.

The brain contains nerve cells that each show maximum activity in response to one particular ITD. One idea is that these nerve cells are arranged in the brain like a map from left to right, and that the brain then uses this map to estimate sound direction. This is known as the Jeffress model, after the scientist who first proposed it. There is some evidence that birds and alligators actually use a system like this to localize sounds, but no such map of nerve cells has yet been identified in mammals. An alternative possibility is that the brain compares activity across groups of ITD-sensitive nerve cells. One of the oldest and simplest ways to measure this is to compare nerve activity in the left and right hemispheres of the brain. This readout is known as the hemispheric difference model.

By analyzing data from published studies, Ihlefeld, Alamatsaz, and Shapley discovered that these two models make opposing predictions about the effects of volume. The Jeffress model predicts that the volume of a sound will not affect a person’s ability to localize it. By contrast, the hemispheric difference model predicts that very soft sounds will lead to systematic errors, so that for the same ITD, softer sounds are perceived closer towards the front than louder sounds. To investigate this further, Ihlefeld, Alamatsaz, and Shapley asked healthy volunteers to localize sounds of different volumes. The volunteers tended to mis-localize quieter sounds, believing them to be closer to the body’s midline than they actually were, which is inconsistent with the predictions of the Jeffress model.

These new findings also reveal key parallels to processing in the visual system. Visual areas of the brain estimate how far away an object is by comparing the input that reaches the two eyes. But these estimates are also systematically less accurate for low-contrast stimuli than for high-contrast ones, just as sound localization is less accurate for softer sounds than for louder ones. The idea that the brain uses the same basic strategy to localize both sights and sounds generates a number of predictions for future studies to test.

A Physiologically Inspired Model for Solving the Cocktail Party Problem

At a cocktail party, we can broadly monitor the entire acoustic scene to detect important cues (e.g., our names being called, or the fire alarm going off), or selectively listen to a target sound source (e.g., a conversation partner). It has recently been observed that individual neurons in the avian field L (analog to the mammalian auditory cortex) can display broad spatial tuning to single targets and selective tuning to a target embedded in spatially distributed sound mixtures. Here, we describe a model inspired by these experimental observations and apply it to process mixtures of human speech sentences. This processing is realized in the neural spiking domain. It converts binaural acoustic inputs into cortical spike trains using a multi-stage model composed of a cochlear filter-bank, a midbrain spatial-localization network, and a cortical network. The output spike trains of the cortical network are then converted back into an acoustic waveform, using a stimulus reconstruction technique. The intelligibility of the reconstructed output is quantified using an objective measure of speech intelligibility. We apply the algorithm to single and multi-talker speech to demonstrate that the physiologically inspired algorithm is able to achieve intelligible reconstruction of an “attended” target sentence embedded in two other non-attended masker sentences. The algorithm is also robust to masker level and displays performance trends comparable to humans. The ideas from this work may help improve the performance of hearing assistive devices (e.g., hearing aids and cochlear implants), speech-recognition technology, and computational algorithms for processing natural scenes cluttered with spatially distributed acoustic objects.

A framework for computational modelling of interaural time difference discrimination of normal and hearing-impaired listeners

Different computational models have been developed to study the interaural time difference (ITD) perception. However, only few have used a physiologically inspired architecture to study ITD discrimination. Furthermore, they do not include aspects of hearing impairment. In this work, a framework was developed to predict ITD thresholds in listeners with normal and impaired hearing. It combines the physiologically inspired model of the auditory periphery proposed by Zilany, Bruce, Nelson, and Carney [(2009). J. Acoust. Soc. Am. 126(5), 2390-2412] as a front end with a coincidence detection stage and a neurometric decision device as a back end. It was validated by comparing its predictions against behavioral data for narrowband stimuli from literature. The framework is able to model ITD discrimination of normal-hearing and hearing-impaired listeners at a group level. Additionally, it was used to explore the effect of different proportions of outer- and inner-hair cell impairment on ITD discrimination.