Deep neural network models reveal interplay of peripheral coding and stimulus statistics in pitch perception

Models optimized for real-world tasks reveal the necessity of precise temporal coding in hearing

Neurons encode information in the timing of their spikes in addition to their firing rates. Spike timing is particularly precise in the auditory nerve, where action potentials phase lock to sound with sub-millisecond precision, but its behavioral relevance is uncertain. To investigate the role of this temporal coding, we optimized machine learning models to perform real-world hearing tasks with simulated cochlear input. We asked how precise auditory nerve spike timing needed to be to reproduce human behavior. Models with high-fidelity phase locking exhibited more human-like sound localization and speech perception than models without, consistent with an essential role in human hearing. Degrading phase locking produced task-dependent effects, revealing how the use of fine-grained temporal information reflects both ecological task demands and neural implementation constraints. The results link neural coding to perception and clarify conditions in which prostheses that fail to restore high-fidelity temporal coding could in principle restore near-normal hearing.

NSF DARE-transforming modeling in neurorehabilitation: a patient-in-the-loop framework

In 2023, the National Science Foundation (NSF) and the National Institute of Health (NIH) brought together engineers, scientists, and clinicians by sponsoring a conference on computational modelling in neurorehabiilitation. To facilitate multidisciplinary collaborations and improve patient care, in this perspective piece we identify where and how computational modelling can support neurorehabilitation. To address the where, we developed a patient-in-the-loop framework that uses multiple and/or continual measurements to update diagnostic and treatment model parameters, treatment type, and treatment prescription, with the goal of maximizing clinically-relevant functional outcomes. This patient-in-the-loop framework has several key features: (i) it includes diagnostic and treatment models, (ii) it is clinically-grounded with the International Classification of Functioning, Disability and Health (ICF) and patient involvement, (iii) it uses multiple or continual data measurements over time, and (iv) it is applicable to a range of neurological and neurodevelopmental conditions. To address the how, we identify state-of-the-art and highlight promising avenues of future research across the realms of sensorimotor adaptation, neuroplasticity, musculoskeletal, and sensory & pain computational modelling. We also discuss both the importance of and how to perform model validation, as well as challenges to overcome when implementing computational models within a clinical setting. The patient-in-the-loop approach offers a unifying framework to guide multidisciplinary collaboration between computational and clinical stakeholders in the field of neurorehabilitation.

Many but not all deep neural network audio models capture brain responses and exhibit correspondence between model stages and brain regions

Models that predict brain responses to stimuli provide one measure of understanding of a sensory system and have many potential applications in science and engineering. Deep artificial neural networks have emerged as the leading such predictive models of the visual system but are less explored in audition. Prior work provided examples of audio-trained neural networks that produced good predictions of auditory cortical fMRI responses and exhibited correspondence between model stages and brain regions, but left it unclear whether these results generalize to other neural network models and, thus, how to further improve models in this domain. We evaluated model-brain correspondence for publicly available audio neural network models along with in-house models trained on 4 different tasks. Most tested models outpredicted standard spectromporal filter-bank models of auditory cortex and exhibited systematic model-brain correspondence: Middle stages best predicted primary auditory cortex, while deep stages best predicted non-primary cortex. However, some state-of-the-art models produced substantially worse brain predictions. Models trained to recognize speech in background noise produced better brain predictions than models trained to recognize speech in quiet, potentially because hearing in noise imposes constraints on biological auditory representations. The training task influenced the prediction quality for specific cortical tuning properties, with best overall predictions resulting from models trained on multiple tasks. The results generally support the promise of deep neural networks as models of audition, though they also indicate that current models do not explain auditory cortical responses in their entirety.

Model metamers reveal divergent invariances between biological and artificial neural networks

Deep neural network models of sensory systems are often proposed to learn representational transformations with invariances like those in the brain. To reveal these invariances, we generated 'model metamers', stimuli whose activations within a model stage are matched to those of a natural stimulus. Metamers for state-of-the-art supervised and unsupervised neural network models of vision and audition were often completely unrecognizable to humans when generated from late model stages, suggesting differences between model and human invariances. Targeted model changes improved human recognizability of model metamers but did not eliminate the overall human-model discrepancy. The human recognizability of a model's metamers was well predicted by their recognizability by other models, suggesting that models contain idiosyncratic invariances in addition to those required by the task. Metamer recognizability dissociated from both traditional brain-based benchmarks and adversarial vulnerability, revealing a distinct failure mode of existing sensory models and providing a complementary benchmark for model assessment.

The neuroconnectionist research programme

Artificial neural networks (ANNs) inspired by biology are beginning to be widely used to model behavioural and neural data, an approach we call 'neuroconnectionism'. ANNs have been not only lauded as the current best models of information processing in the brain but also criticized for failing to account for basic cognitive functions. In this Perspective article, we propose that arguing about the successes and failures of a restricted set of current ANNs is the wrong approach to assess the promise of neuroconnectionism for brain science. Instead, we take inspiration from the philosophy of science, and in particular from Lakatos, who showed that the core of a scientific research programme is often not directly falsifiable but should be assessed by its capacity to generate novel insights. Following this view, we present neuroconnectionism as a general research programme centred around ANNs as a computational language for expressing falsifiable theories about brain computation. We describe the core of the programme, the underlying computational framework and its tools for testing specific neuroscientific hypotheses and deriving novel understanding. Taking a longitudinal view, we review past and present neuroconnectionist projects and their responses to challenges and argue that the research programme is highly progressive, generating new and otherwise unreachable insights into the workings of the brain.

Sensitivity to Frequency Modulation is Limited Centrally

Modulations in both amplitude and frequency are prevalent in natural sounds and are critical in defining their properties. Humans are exquisitely sensitive to frequency modulation (FM) at the slow modulation rates and low carrier frequencies that are common in speech and music. This enhanced sensitivity to slow-rate and low-frequency FM has been widely believed to reflect precise, stimulus-driven phase locking to temporal fine structure in the auditory nerve. At faster modulation rates and/or higher carrier frequencies, FM is instead thought to be coded by coarser frequency-to-place mapping, where FM is converted to amplitude modulation (AM) via cochlear filtering. Here, we show that patterns of human FM perception that have classically been explained by limits in peripheral temporal coding are instead better accounted for by constraints in the central processing of fundamental frequency (F0) or pitch. We measured FM detection in male and female humans using harmonic complex tones with an F0 within the range of musical pitch but with resolved harmonic components that were all above the putative limits of temporal phase locking (>8 kHz). Listeners were more sensitive to slow than fast FM rates, even though all components were beyond the limits of phase locking. In contrast, AM sensitivity remained better at faster than slower rates, regardless of carrier frequency. These findings demonstrate that classic trends in human FM sensitivity, previously attributed to auditory nerve phase locking, may instead reflect the constraints of a unitary code that operates at a more central level of processing.SIGNIFICANCE STATEMENT Natural sounds involve dynamic frequency and amplitude fluctuations. Humans are particularly sensitive to frequency modulation (FM) at slow rates and low carrier frequencies, which are prevalent in speech and music. This sensitivity has been ascribed to encoding of stimulus temporal fine structure (TFS) via phase-locked auditory nerve activity. To test this long-standing theory, we measured FM sensitivity using complex tones with a low F0 but only high-frequency harmonics beyond the limits of phase locking. Dissociating the F0 from TFS showed that FM sensitivity is limited not by peripheral encoding of TFS but rather by central processing of F0, or pitch. The results suggest a unitary code for FM detection limited by more central constraints.

Encoding of speech in convolutional layers and the brain stem based on language experience

Comparing artificial neural networks with outputs of neuroimaging techniques has recently seen substantial advances in (computer) vision and text-based language models. Here, we propose a framework to compare biological and artificial neural computations of spoken language representations and propose several new challenges to this paradigm. The proposed technique is based on a similar principle that underlies electroencephalography (EEG): averaging of neural (artificial or biological) activity across neurons in the time domain, and allows to compare encoding of any acoustic property in the brain and in intermediate convolutional layers of an artificial neural network. Our approach allows a direct comparison of responses to a phonetic property in the brain and in deep neural networks that requires no linear transformations between the signals. We argue that the brain stem response (cABR) and the response in intermediate convolutional layers to the exact same stimulus are highly similar without applying any transformations, and we quantify this observation. The proposed technique not only reveals similarities, but also allows for analysis of the encoding of actual acoustic properties in the two signals: we compare peak latency (i) in cABR relative to the stimulus in the brain stem and in (ii) intermediate convolutional layers relative to the input/output in deep convolutional networks. We also examine and compare the effect of prior language exposure on the peak latency in cABR and in intermediate convolutional layers. Substantial similarities in peak latency encoding between the human brain and intermediate convolutional networks emerge based on results from eight trained networks (including a replication experiment). The proposed technique can be used to compare encoding between the human brain and intermediate convolutional layers for any acoustic property and for other neuroimaging techniques.

Quantitative models of auditory cortical processing

To generate insight from experimental data, it is critical to understand the inter-relationships between individual data points and place them in context within a structured framework. Quantitative modeling can provide the scaffolding for such an endeavor. Our main objective in this review is to provide a primer on the range of quantitative tools available to experimental auditory neuroscientists. Quantitative modeling is advantageous because it can provide a compact summary of observed data, make underlying assumptions explicit, and generate predictions for future experiments. Quantitative models may be developed to characterize or fit observed data, to test theories of how a task may be solved by neural circuits, to determine how observed biophysical details might contribute to measured activity patterns, or to predict how an experimental manipulation would affect neural activity. In complexity, quantitative models can range from those that are highly biophysically realistic and that include detailed simulations at the level of individual synapses, to those that use abstract and simplified neuron models to simulate entire networks. Here, we survey the landscape of recently developed models of auditory cortical processing, highlighting a small selection of models to demonstrate how they help generate insight into the mechanisms of auditory processing. We discuss examples ranging from models that use details of synaptic properties to explain the temporal pattern of cortical responses to those that use modern deep neural networks to gain insight into human fMRI data. We conclude by discussing a biologically realistic and interpretable model that our laboratory has developed to explore aspects of vocalization categorization in the auditory pathway.

Using artificial neural networks to ask 'why' questions of minds and brains

Neuroscientists have long characterized the properties and functions of the nervous system, and are increasingly succeeding in answering how brains perform the tasks they do. But the question 'why' brains work the way they do is asked less often. The new ability to optimize artificial neural networks (ANNs) for performance on human-like tasks now enables us to approach these 'why' questions by asking when the properties of networks optimized for a given task mirror the behavioral and neural characteristics of humans performing the same task. Here we highlight the recent success of this strategy in explaining why the visual and auditory systems work the way they do, at both behavioral and neural levels.

A unitary model of auditory frequency change perception

Changes in the frequency content of sounds over time are arguably the most basic form of information about the behavior of sound-emitting objects. In perceptual studies, such changes have mostly been investigated separately, as aspects of either pitch or timbre. Here, we propose a unitary account of "up" and "down" subjective judgments of frequency change, based on a model combining auditory correlates of acoustic cues in a sound-specific and listener-specific manner. To do so, we introduce a generalized version of so-called Shepard tones, allowing symmetric manipulations of spectral information on a fine scale, usually associated to pitch (spectral fine structure, SFS), and on a coarse scale, usually associated timbre (spectral envelope, SE). In a series of behavioral experiments, listeners reported "up" or "down" shifts across pairs of generalized Shepard tones that differed in SFS, in SE, or in both. We observed the classic properties of Shepard tones for either SFS or SE shifts: subjective judgements followed the smallest log-frequency change direction, with cases of ambiguity and circularity. Interestingly, when both SFS and SE changes were applied concurrently (synergistically or antagonistically), we observed a trade-off between cues. Listeners were encouraged to report when they perceived "both" directions of change concurrently, but this rarely happened, suggesting a unitary percept. A computational model could accurately fit the behavioral data by combining different cues reflecting frequency changes after auditory filtering. The model revealed that cue weighting depended on the nature of the sound. When presented with harmonic sounds, listeners put more weight on SFS-related cues, whereas inharmonic sounds led to more weight on SE-related cues. Moreover, these stimulus-based factors were modulated by inter-individual differences, revealing variability across listeners in the detailed recipe for "up" and "down" judgments. We argue that frequency changes are tracked perceptually via the adaptive combination of a diverse set of cues, in a manner that is in fact similar to the derivation of other basic auditory dimensions such as spatial location.

Questions and controversies surrounding the perception and neural coding of pitch

Pitch is a fundamental aspect of auditory perception that plays an important role in our ability to understand speech, appreciate music, and attend to one sound while ignoring others. The questions surrounding how pitch is represented in the auditory system, and how our percept relates to the underlying acoustic waveform, have been a topic of inquiry and debate for well over a century. New findings and technological innovations have led to challenges of some long-standing assumptions and have raised new questions. This article reviews some recent developments in the study of pitch coding and perception and focuses on the topic of how pitch information is extracted from peripheral representations based on frequency-to-place mapping (tonotopy), stimulus-driven auditory-nerve spike timing (phase locking), or a combination of both. Although a definitive resolution has proved elusive, the answers to these questions have potentially important implications for mitigating the effects of hearing loss via devices such as cochlear implants.

Cochlear tuning and the peripheral representation of harmonic sounds in mammals

Albert Feng was a prominent comparative neurophysiologist whose research provided numerous contributions towards understanding how the spectral and temporal characteristics of vocalizations underlie sound communication in frogs and bats. The present study is dedicated to Al's memory and compares the spectral and temporal representations of stochastic, complex sounds which underlie the perception of pitch strength in humans and chinchillas. Specifically, the pitch strengths of these stochastic sounds differ between humans and chinchillas, suggesting that humans and chinchillas may be using different cues. Outputs of auditory filterbank models based on human and chinchilla cochlear tuning were examined. Excitation patterns of harmonics are enhanced in humans as compared with chinchillas. In contrast, summary correlograms are degraded in humans as compared with chinchillas. Comparing summary correlograms and excitation patterns with corresponding behavioral data on pitch strength suggests that the dominant cue for pitch strength in humans is spectral (i.e., harmonic) structure, whereas the dominant cue for chinchillas is temporal (i.e., envelope) structure. The results support arguments that the broader cochlear tuning in non-human mammals emphasizes temporal cues for pitch perception, whereas the sharper cochlear tuning in humans emphasizes spectral cues.

Human Auditory Ecology: Extending Hearing Research to the Perception of Natural Soundscapes by Humans in Rapidly Changing Environments

Research in hearing sciences has provided extensive knowledge about how the human auditory system processes speech and assists communication. In contrast, little is known about how this system processes "natural soundscapes," that is the complex arrangements of biological and geophysical sounds shaped by sound propagation through non-anthropogenic habitats [Grinfeder et al. (2022). Frontiers in Ecology and Evolution. 10: 894232]. This is surprising given that, for many species, the capacity to process natural soundscapes determines survival and reproduction through the ability to represent and monitor the immediate environment. Here we propose a framework to encourage research programmes in the field of "human auditory ecology," focusing on the study of human auditory perception of ecological processes at work in natural habitats. Based on large acoustic databases with high ecological validity, these programmes should investigate the extent to which this presumably ancestral monitoring function of the human auditory system is adapted to specific information conveyed by natural soundscapes, whether it operate throughout the life span or whether it emerges through individual learning or cultural transmission. Beyond fundamental knowledge of human hearing, these programmes should yield a better understanding of how normal-hearing and hearing-impaired listeners monitor rural and city green and blue spaces and benefit from them, and whether rehabilitation devices (hearing aids and cochlear implants) restore natural soundscape perception and emotional responses back to normal. Importantly, they should also reveal whether and how humans hear the rapid changes in the environment brought about by human activity.

Periodicity Pitch Perception Part III: Sensibility and Pachinko Volatility

Neuromorphic computer models are used to explain sensory perceptions. Auditory models generate cochleagrams, which resemble the spike distributions in the auditory nerve. Neuron ensembles along the auditory pathway transform sensory inputs step by step and at the end pitch is represented in auditory categorical spaces. In two previous articles in the series on periodicity pitch perception an extended auditory model had been successfully used for explaining periodicity pitch proved for various musical instrument generated tones and sung vowels. In this third part in the series the focus is on octopus cells as they are central sensitivity elements in auditory cognition processes. A powerful numerical model had been devised, in which auditory nerve fibers (ANFs) spike events are the inputs, triggering the impulse responses of the octopus cells. Efficient algorithms are developed and demonstrated to explain the behavior of octopus cells with a focus on a simple event-based hardware implementation of a layer of octopus neurons. The main finding is, that an octopus' cell model in a local receptive field fine-tunes to a specific trajectory by a spike-timing-dependent plasticity (STDP) learning rule with synaptic pre-activation and the dendritic back-propagating signal as post condition. Successful learning explains away the teacher and there is thus no need for a temporally precise control of plasticity that distinguishes between learning and retrieval phases. Pitch learning is cascaded: At first octopus cells respond individually by self-adjustment to specific trajectories in their local receptive fields, then unions of octopus cells are collectively learned for pitch discrimination. Pitch estimation by inter-spike intervals is shown exemplary using two input scenarios: a simple sinus tone and a sung vowel. The model evaluation indicates an improvement in pitch estimation on a fixed time-scale.

Human discrimination and modeling of high-frequency complex tones shed light on the neural codes for pitch

Accurate pitch perception of harmonic complex tones is widely believed to rely on temporal fine structure information conveyed by the precise phase-locked responses of auditory-nerve fibers. However, accurate pitch perception remains possible even when spectrally resolved harmonics are presented at frequencies beyond the putative limits of neural phase locking, and it is unclear whether residual temporal information, or a coarser rate-place code, underlies this ability. We addressed this question by measuring human pitch discrimination at low and high frequencies for harmonic complex tones, presented either in isolation or in the presence of concurrent complex-tone maskers. We found that concurrent complex-tone maskers impaired performance at both low and high frequencies, although the impairment introduced by adding maskers at high frequencies relative to low frequencies differed between the tested masker types. We then combined simulated auditory-nerve responses to our stimuli with ideal-observer analysis to quantify the extent to which performance was limited by peripheral factors. We found that the worsening of both frequency discrimination and F0 discrimination at high frequencies could be well accounted for (in relative terms) by optimal decoding of all available information at the level of the auditory nerve. A Python package is provided to reproduce these results, and to simulate responses to acoustic stimuli from the three previously published models of the human auditory nerve used in our analyses.

Making sense of periodicity glimpses in a prediction-update-loop-A computational model of attentive voice tracking

Humans are able to follow a speaker even in challenging acoustic conditions. The perceptual mechanisms underlying this ability remain unclear. A computational model of attentive voice tracking, consisting of four computational blocks: (1) sparse periodicity-based auditory features (sPAF) extraction, (2) foreground-background segregation, (3) state estimation, and (4) top-down knowledge, is presented. The model connects the theories about auditory glimpses, foreground-background segregation, and Bayesian inference. It is implemented with the sPAF, sequential Monte Carlo sampling, and probabilistic voice models. The model is evaluated by comparing it with the human data obtained in the study by Woods and McDermott [Curr. Biol. 25(17), 2238-2246 (2015)], which measured the ability to track one of two competing voices with time-varying parameters [fundamental frequency (F0) and formants (F1,F2)]. Three model versions were tested, which differ in the type of information used for the segregation: version (a) uses the oracle F0, version (b) uses the estimated F0, and version (c) uses the spectral shape derived from the estimated F0 and oracle F1 and F2. Version (a) simulates the optimal human performance in conditions with the largest separation between the voices, version (b) simulates the conditions in which the separation in not sufficient to follow the voices, and version (c) is closest to the human performance for moderate voice separation.