Audio-visual synchrony modulates the ventriloquist illusion and its neural/spatial representation in the auditory cortex

Perceptual training of audiovisual simultaneity judgments generalizes across spatial locations

Multisensory processing critically depends on the perceived timing of stimuli in the different sensory modalities. Crossmodal stimuli that fall within rather than outside an individual temporal binding window (TBW) are more likely to be bound into a multisensory percept. A number of studies have shown that a short perceptual training in which participants receive feedback on their responses in an audiovisual simultaneity judgment (SJ) task can substantially decrease the size of the TBW and hence increase crossmodal temporal acuity. Here we tested whether multisensory perceptual learning in the SJ task is specific for the spatial locations at which the audiovisual stimuli are presented during training. Participants received feedback about the correctness of their SJ responses for audiovisual stimuli which were presented in one hemifield only. The TBW was assessed separately for audiovisual stimuli in each hemifield before and one day after the training. In line with previous findings, the size of the TBW was significantly reduced after the training phase. Importantly, an equally strong reduction of TBW size was observed in both the trained and the untrained hemifield. Thus, multisensory temporal learning completely generalized to the untrained hemifield, suggesting that the improvement in crossmodal temporal acuity was mediated by higher, location-invariant processing stages. These findings have implications for the design of multisensory training protocols in applied settings such as clinical interventions by showing that training at multiple spatial locations might not be necessary to achieve robust improvements in crossmodal temporal acuity.

Integration of somatosensory and motor-related information in the auditory system

An ability to integrate information provided by different sensory modalities is a fundamental feature of neurons in many brain areas. Because visual and auditory inputs often originate from the same external object, which may be located some distance away from the observer, the synthesis of these cues can improve localization accuracy and speed up behavioral responses. By contrast, multisensory interactions occurring close to the body typically involve a combination of tactile stimuli with other sensory modalities. Moreover, most activities involving active touch generate sound, indicating that stimuli in these modalities are frequently experienced together. In this review, we examine the basis for determining sound-source distance and the contribution of auditory inputs to the neural encoding of space around the body. We then consider the perceptual consequences of combining auditory and tactile inputs in humans and discuss recent evidence from animal studies demonstrating how cortical and subcortical areas work together to mediate communication between these senses. This research has shown that somatosensory inputs interface with and modulate sound processing at multiple levels of the auditory pathway, from the cochlear nucleus in the brainstem to the cortex. Circuits involving inputs from the primary somatosensory cortex to the auditory midbrain have been identified that mediate suppressive effects of whisker stimulation on auditory thalamocortical processing, providing a possible basis for prioritizing the processing of tactile cues from nearby objects. Close links also exist between audition and movement, and auditory responses are typically suppressed by locomotion and other actions. These movement-related signals are thought to cancel out self-generated sounds, but they may also affect auditory responses via the associated somatosensory stimulation or as a result of changes in brain state. Together, these studies highlight the importance of considering both multisensory context and movement-related activity in order to understand how the auditory cortex operates during natural behaviors, paving the way for future work to investigate auditory-somatosensory interactions in more ecological situations.

Meta-Analyses Support a Taxonomic Model for Representations of Different Categories of Audio-Visual Interaction Events in the Human Brain

Our ability to perceive meaningful action events involving objects, people, and other animate agents is characterized in part by an interplay of visual and auditory sensory processing and their cross-modal interactions. However, this multisensory ability can be altered or dysfunctional in some hearing and sighted individuals, and in some clinical populations. The present meta-analysis sought to test current hypotheses regarding neurobiological architectures that may mediate audio-visual multisensory processing. Reported coordinates from 82 neuroimaging studies (137 experiments) that revealed some form of audio-visual interaction in discrete brain regions were compiled, converted to a common coordinate space, and then organized along specific categorical dimensions to generate activation likelihood estimate (ALE) brain maps and various contrasts of those derived maps. The results revealed brain regions (cortical "hubs") preferentially involved in multisensory processing along different stimulus category dimensions, including 1) living versus nonliving audio-visual events, 2) audio-visual events involving vocalizations versus actions by living sources, 3) emotionally valent events, and 4) dynamic-visual versus static-visual audio-visual stimuli. These meta-analysis results are discussed in the context of neurocomputational theories of semantic knowledge representations and perception, and the brain volumes of interest are available for download to facilitate data interpretation for future neuroimaging studies.

Re-weighting of Sound Localization Cues by Audiovisual Training

Sound localization requires the integration in the brain of auditory spatial cues generated by interactions with the external ears, head and body. Perceptual learning studies have shown that the relative weighting of these cues can change in a context-dependent fashion if their relative reliability is altered. One factor that may influence this process is vision, which tends to dominate localization judgments when both modalities are present and induces a recalibration of auditory space if they become misaligned. It is not known, however, whether vision can alter the weighting of individual auditory localization cues. Using virtual acoustic space stimuli, we measured changes in subjects’ sound localization biases and binaural localization cue weights after ∼50 min of training on audiovisual tasks in which visual stimuli were either informative or not about the location of broadband sounds. Four different spatial configurations were used in which we varied the relative reliability of the binaural cues: interaural time differences (ITDs) and frequency-dependent interaural level differences (ILDs). In most subjects and experiments, ILDs were weighted more highly than ITDs before training. When visual cues were spatially uninformative, some subjects showed a reduction in auditory localization bias and the relative weighting of ILDs increased after training with congruent binaural cues. ILDs were also upweighted if they were paired with spatially-congruent visual cues, and the largest group-level improvements in sound localization accuracy occurred when both binaural cues were matched to visual stimuli. These data suggest that binaural cue reweighting reflects baseline differences in the relative weights of ILDs and ITDs, but is also shaped by the availability of congruent visual stimuli. Training subjects with consistently misaligned binaural and visual cues produced the ventriloquism aftereffect, i.e., a corresponding shift in auditory localization bias, without affecting the inter-subject variability in sound localization judgments or their binaural cue weights. Our results show that the relative weighting of different auditory localization cues can be changed by training in ways that depend on their reliability as well as the availability of visual spatial information, with the largest improvements in sound localization likely to result from training with fully congruent audiovisual information.

The Effects of Audiovisual Inputs on Solving the Cocktail Party Problem in the Human Brain: An fMRI Study

At cocktail parties, our brains often simultaneously receive visual and auditory information. Although the cocktail party problem has been widely investigated under auditory-only settings, the effects of audiovisual inputs have not. This study explored the effects of audiovisual inputs in a simulated cocktail party. In our fMRI experiment, each congruent audiovisual stimulus was a synthesis of 2 facial movie clips, each of which could be classified into 1 of 2 emotion categories (crying and laughing). Visual-only (faces) and auditory-only stimuli (voices) were created by extracting the visual and auditory contents from the synthesized audiovisual stimuli. Subjects were instructed to selectively attend to 1 of the 2 objects contained in each stimulus and to judge its emotion category in the visual-only, auditory-only, and audiovisual conditions. The neural representations of the emotion features were assessed by calculating decoding accuracy and brain pattern-related reproducibility index based on the fMRI data. We compared the audiovisual condition with the visual-only and auditory-only conditions and found that audiovisual inputs enhanced the neural representations of emotion features of the attended objects instead of the unattended objects. This enhancement might partially explain the benefits of audiovisual inputs for the brain to solve the cocktail party problem.

The Ventriloquist Illusion as a Tool to Study Multisensory Processing: An Update

Ventriloquism, the illusion that a voice appears to come from the moving mouth of a puppet rather than from the actual speaker, is one of the classic examples of multisensory processing. In the laboratory, this illusion can be reliably induced by presenting simple meaningless audiovisual stimuli with a spatial discrepancy between the auditory and visual components. Typically, the perceived location of the sound source is biased toward the location of the visual stimulus (the ventriloquism effect). The strength of the visual bias reflects the relative reliability of the visual and auditory inputs as well as prior expectations that the two stimuli originated from the same source. In addition to the ventriloquist illusion, exposure to spatially discrepant audiovisual stimuli results in a subsequent recalibration of unisensory auditory localization (the ventriloquism aftereffect). In the past years, the ventriloquism effect and aftereffect have seen a resurgence as an experimental tool to elucidate basic mechanisms of multisensory integration and learning. For example, recent studies have: (a) revealed top-down influences from the reward and motor systems on cross-modal binding; (b) dissociated recalibration processes operating at different time scales; and (c) identified brain networks involved in the neuronal computations underlying multisensory integration and learning. This mini review article provides a brief overview of established experimental paradigms to measure the ventriloquism effect and aftereffect before summarizing these pathbreaking new advancements. Finally, it is pointed out how the ventriloquism effect and aftereffect could be utilized to address some of the current open questions in the field of multisensory research.

Shared neural underpinnings of multisensory integration and trial-by-trial perceptual recalibration in humans

Perception adapts to mismatching multisensory information, both when different cues appear simultaneously and when they appear sequentially. While both multisensory integration and adaptive trial-by-trial recalibration are central for behavior, it remains unknown whether they are mechanistically linked and arise from a common neural substrate. To relate the neural underpinnings of sensory integration and recalibration, we measured whole-brain magnetoencephalography while human participants performed an audio-visual ventriloquist task. Using single-trial multivariate analysis, we localized the perceptually-relevant encoding of multisensory information within and between trials. While we found neural signatures of multisensory integration within temporal and parietal regions, only medial superior parietal activity encoded past and current sensory information and mediated the perceptual recalibration within and between trials. These results highlight a common neural substrate of sensory integration and perceptual recalibration, and reveal a role of medial parietal regions in linking present and previous multisensory evidence to guide adaptive behavior.

A good ventriloquist will make their audience experience an illusion. The speech the spectators hear appears to come from the mouth of the puppet and not from the puppeteer. Moviegoers experience the same illusion: they perceive dialogue as coming from the mouths of the actors on screen, rather than from the loudspeakers mounted on the walls. Known as the ventriloquist effect, this ‘trick’ exists because the brain assumes that sights and sounds which occur at the same time have the same origin, and it therefore combines the two sets of sensory stimuli.

A version of the ventriloquist effect can be induced in the laboratory. Participants hear a sound while watching a simple visual stimulus (for instance, a circle) appear on a screen. When asked to pinpoint the origin of the noise, volunteers choose a location shifted towards the circle, even if this was not where the sound came from. In addition, this error persists when the visual stimulus is no longer present: if a standard trial is followed by a trial that features a sound but no circle, participants perceive the sound in the second test as ‘drawn’ towards the direction of the former shift. This is known as the ventriloquist aftereffect.

By scanning the brains of healthy volunteers performing this task, Park and Kayser show that a number of brain areas contribute to the ventriloquist effect. All of these regions help to combine what we see with what we hear, but only one maintains representations of the combined sensory inputs over time. Called the medial superior parietal cortex, this area is unique in contributing to both the ventriloquist effect and its aftereffect.

We must constantly use past and current sensory information to adapt our behavior to the environment. The results by Park and Kayser shed light on the brain structures that underpin our capacity to combine information from several senses, as well as our ability to encode memories. Such knowledge should be useful to explore how we can make flexible decisions.

Reduced multisensory integration of self-initiated stimuli

The processing and perception of stimuli is altered when these stimuli are not passively presented but rather are actively triggered, or "self-initiated", by the participants. For unimodal stimuli, perceptual changes in stimulus timing and intensity have been demonstrated. Initial results have suggested that self-initiation may affect multisensory processing as well. The present study examined the effects of self-initiation on audiovisual integration in the ventriloquism effect (VE), that is, the mislocalization of auditory stimuli toward a spatially displaced visual stimulus. The effects of self-initiation on the VE were investigated with audiovisual stimuli that featured varying degrees of spatial and temporal separation. Stimuli were either triggered by the participants' button press or not, and stimulus onsets were either predictable or not. Arguing from the perspective of Bayesian causal inference models, we hypothesized self-initiation to increase the prior probability of two stimuli being integrated. Contrary to this intuitive assumption, less VE was observed when the stimuli were self-initiated by the participants than when they were externally generated. Since no effects of self-initiation on unimodal processing were observed, these effects must specifically pertain to multisensory processes. Finally, data were fit with a causal inference model, where self-initiation was associated with a reduction of the prior probability to integrate audiovisual stimuli. In conclusion, the presence of a self-initiated motor signal influences audiovisual integration, such that auditory localization is less biased by visual stimuli, which likely depends on top-down signals.

Spatial receptive field shift by preceding cross-modal stimulation in the cat superior colliculus

KEY POINTS

It has been known for some time that sensory information of one type can bias the spatial perception of another modality. However, there is a lack of evidence of this occurring in individual neurons. In the present study, we found that the spatial receptive field of superior colliculus multisensory neurons could be dynamically shifted by a preceding stimulus in a different modality. The extent to which the receptive field shifted was dependent on both temporal and spatial gaps between the preceding and following stimuli, as well as the salience of the preceding stimulus. This result provides a neural mechanism that could underlie the process of cross-modal spatial calibration.

ABSTRACT

Psychophysical studies have shown that the different senses can be spatially entrained by each other. This can be observed in certain phenomena, such as ventriloquism, in which a visual stimulus can attract the perceived location of a spatially discordant sound. However, the neural mechanism underlying this cross-modal spatial recalibration has remained unclear, as has whether it takes place dynamically. We explored these issues in multisensory neurons of the cat superior colliculus (SC), a midbrain structure that involves both cross-modal and sensorimotor integration. Sequential cross-modal stimulation showed that the preceding stimulus can shift the receptive field (RF) of the lagging response. This cross-modal spatial calibration took place in both auditory and visual RFs, although auditory RFs shifted slightly more. By contrast, if a preceding stimulus was from the same modality, it failed to induce a similarly substantial RF shift. The extent of the RF shift was dependent on both temporal and spatial gaps between the preceding and following stimuli, as well as the salience of the preceding stimulus. A narrow time gap and high stimulus salience were able to induce larger RF shifts. In addition, when both visual and auditory stimuli were presented simultaneously, a substantial RF shift toward the location-fixed stimulus was also induced. These results, taken together, reveal an online cross-modal process and reflect the details of the organization of SC inter-sensory spatial calibration.

Invisible Flashes Alter Perceived Sound Location

Information integration across the senses is fundamental for effective interactions with our environment. The extent to which signals from different senses can interact in the absence of awareness is controversial. Combining the spatial ventriloquist illusion and dynamic continuous flash suppression (dCFS), we investigated in a series of two experiments whether visual signals that observers do not consciously perceive can influence spatial perception of sounds. Importantly, dCFS obliterated visual awareness only on a fraction of trials allowing us to compare spatial ventriloquism for physically identical flashes that were judged as visible or invisible. Our results show a stronger ventriloquist effect for visible than invisible flashes. Critically, a robust ventriloquist effect emerged also for invisible flashes even when participants were at chance when locating the flash. Collectively, our findings demonstrate that signals that we are not aware of in one sensory modality can alter spatial perception of signals in another sensory modality.

Causal inference and temporal predictions in audiovisual perception of speech and music

To form a coherent percept of the environment, the brain must integrate sensory signals emanating from a common source but segregate those from different sources. Temporal regularities are prominent cues for multisensory integration, particularly for speech and music perception. In line with models of predictive coding, we suggest that the brain adapts an internal model to the statistical regularities in its environment. This internal model enables cross-sensory and sensorimotor temporal predictions as a mechanism to arbitrate between integration and segregation of signals from different senses.

Reliability-Weighted Integration of Audiovisual Signals Can Be Modulated by Top-down Attention

Behaviorally, it is well established that human observers integrate signals near-optimally weighted in proportion to their reliabilities as predicted by maximum likelihood estimation. Yet, despite abundant behavioral evidence, it is unclear how the human brain accomplishes this feat. In a spatial ventriloquist paradigm, participants were presented with auditory, visual, and audiovisual signals and reported the location of the auditory or the visual signal. Combining psychophysics, multivariate functional MRI (fMRI) decoding, and models of maximum likelihood estimation (MLE), we characterized the computational operations underlying audiovisual integration at distinct cortical levels. We estimated observers' behavioral weights by fitting psychometric functions to participants' localization responses. Likewise, we estimated the neural weights by fitting neurometric functions to spatial locations decoded from regional fMRI activation patterns. Our results demonstrate that low-level auditory and visual areas encode predominantly the spatial location of the signal component of a region's preferred auditory (or visual) modality. By contrast, intraparietal sulcus forms spatial representations by integrating auditory and visual signals weighted by their reliabilities. Critically, the neural and behavioral weights and the variance of the spatial representations depended not only on the sensory reliabilities as predicted by the MLE model but also on participants' modality-specific attention and report (i.e., visual vs. auditory). These results suggest that audiovisual integration is not exclusively determined by bottom-up sensory reliabilities. Instead, modality-specific attention and report can flexibly modulate how intraparietal sulcus integrates sensory signals into spatial representations to guide behavioral responses (e.g., localization and orienting).

Spatial and frequency specificity of the ventriloquism aftereffect revisited

Exposure to audiovisual stimuli with a consistent spatial misalignment seems to result in a recalibration of unisensory auditory spatial representations. The previous studies have suggested that this so-called ventriloquism aftereffect is confined to the trained region of space, but yielded inconsistent results as to whether or not recalibration generalizes to untrained sound frequencies. Here, we reassessed the spatial and frequency specificity of the ventriloquism aftereffect by testing whether auditory spatial perception can be independently recalibrated for two different sound frequencies and/or at two different spatial locations. Recalibration was confined to locations within the trained hemifield, suggesting that spatial representations were independently adjusted for the two hemifields. The frequency specificity of the ventriloquism aftereffect depended on the presence or the absence of conflicting audiovisual adaptation stimuli within the same hemifield. Moreover, adaptation of two different sound frequencies in opposite directions (leftward vs. rightward) resulted in a selective suppression of leftward recalibration, even when the adapting stimuli were presented in different hemifields. Thus, instead of representing a fixed stimulus-driven process, cross-modal recalibration seems to critically depend on the sensory context and takes into account inconsistencies in the cross-modal input.

Supramodal effect of rightward prismatic adaptation on spatial representations within the ventral attentional system

Rightward prismatic adaptation (R-PA) was shown to alleviate not only visuo-spatial but also auditory symptoms in neglect. The neural mechanisms underlying the effect of R-PA have been previously investigated in visual tasks, demonstrating a shift of hemispheric dominance for visuo-spatial attention from the right to the left hemisphere both in normal subjects and in patients. We have investigated whether the same neural mechanisms underlie the supramodal effect of R-PA on auditory attention. Normal subjects underwent a brief session of R-PA, which was preceded and followed by an fMRI evaluation during which subjects detected targets within the left, central and right space in the auditory or visual modality. R-PA-related changes in activation patterns were found bilaterally in the inferior parietal lobule. In either modality, the representation of the left, central and right space increased in the left IPL, whereas the representation of the right space decreased in the right IPL. Thus, a brief exposure to R-PA modulated the representation of the auditory and visual space within the ventral attentional system. This shift in hemispheric dominance for auditory spatial attention offers a parsimonious explanation for the previously reported effects of R-PA on auditory symptoms in neglect.

For Better or Worse: The Effect of Prismatic Adaptation on Auditory Neglect

Patients with auditory neglect attend less to auditory stimuli on their left and/or make systematic directional errors when indicating sound positions. Rightward prismatic adaptation (R-PA) was repeatedly shown to alleviate symptoms of visuospatial neglect and once to restore partially spatial bias in dichotic listening. It is currently unknown whether R-PA affects only this ear-related symptom or also other aspects of auditory neglect. We have investigated the effect of R-PA on left ear extinction in dichotic listening, space-related inattention assessed by diotic listening, and directional errors in auditory localization in patients with auditory neglect. The most striking effect of R-PA was the alleviation of left ear extinction in dichotic listening, which occurred in half of the patients with initial deficit. In contrast to nonresponders, their lesions spared the right dorsal attentional system and posterior temporal cortex. The beneficial effect of R-PA on an ear-related performance contrasted with detrimental effects on diotic listening and auditory localization. The former can be parsimoniously explained by the SHD-VAS model (shift in hemispheric dominance within the ventral attentional system; Clarke and Crottaz-Herbette 2016), which is based on the R-PA-induced shift of the right-dominant ventral attentional system to the left hemisphere. The negative effects in space-related tasks may be due to the complex nature of auditory space encoding at a cortical level.

The role of auditory cortex in the spatial ventriloquism aftereffect

Cross-modal recalibration allows the brain to maintain coherent sensory representations of the world. Using functional magnetic resonance imaging (fMRI), the present study aimed at identifying the neural mechanisms underlying recalibration in an audiovisual ventriloquism aftereffect paradigm. Participants performed a unimodal sound localization task, before and after they were exposed to adaptation blocks, in which sounds were paired with spatially disparate visual stimuli offset by 14° to the right. Behavioral results showed a significant rightward shift in sound localization following adaptation, indicating a ventriloquism aftereffect. Regarding fMRI results, left and right planum temporale (lPT/rPT) were found to respond more to contralateral sounds than to central sounds at pretest. Contrasting posttest with pretest blocks revealed significantly enhanced fMRI-signals in space-sensitive lPT after adaptation, matching the behavioral rightward shift in sound localization. Moreover, a region-of-interest analysis in lPT/rPT revealed that the lPT activity correlated positively with the localization shift for right-side sounds, whereas rPT activity correlated negatively with the localization shift for left-side and central sounds. Finally, using functional connectivity analysis, we observed enhanced coupling of the lPT with left and right inferior parietal areas as well as left motor regions following adaptation and a decoupling of lPT/rPT with contralateral auditory cortex, which scaled with participants' degree of adaptation. Together, the fMRI results suggest that cross-modal spatial recalibration is accomplished by an adjustment of unisensory representations in low-level auditory cortex. Such persistent adjustments of low-level sensory representations seem to be mediated by the interplay with higher-level spatial representations in parietal cortex.

Accumulation and decay of visual capture and the ventriloquism aftereffect caused by brief audio-visual disparities

Visual capture and the ventriloquism aftereffect resolve spatial disparities of incongruent auditory visual (AV) objects by shifting auditory spatial perception to align with vision. Here, we demonstrated the distinct temporal characteristics of visual capture and the ventriloquism aftereffect in response to brief AV disparities. In a set of experiments, subjects localized either the auditory component of AV targets (A within AV) or a second sound presented at varying delays (1-20 s) after AV exposure (A2 after AV). AV targets were trains of brief presentations (1 or 20), covering a ±30° azimuthal range, and with ±8° (R or L) disparity. We found that the magnitude of visual capture generally reached its peak within a single AV pair and did not dissipate with time, while the ventriloquism aftereffect accumulated with repetitions of AV pairs and dissipated with time. Additionally, the magnitude of the auditory shift induced by each phenomenon was uncorrelated across listeners and visual capture was unaffected by subsequent auditory targets, indicating that visual capture and the ventriloquism aftereffect are separate mechanisms with distinct effects on auditory spatial perception. Our results indicate that visual capture is a 'sample-and-hold' process that binds related objects and stores the combined percept in memory, whereas the ventriloquism aftereffect is a 'leaky integrator' process that accumulates with experience and decays with time to compensate for cross-modal disparities.

Cortical Correlates of Low-Level Perception: From Neural Circuits to Percepts

Low-level perception results from neural-based computations, which build a multimodal skeleton of unconscious or self-generated inferences on our environment. This review identifies bottleneck issues concerning the role of early primary sensory cortical areas, mostly in rodent and higher mammals (cats and non-human primates), where perception substrates can be searched at multiple scales of neural integration. We discuss the limitation of purely bottom-up approaches for providing realistic models of early sensory processing and the need for identification of fast adaptive processes, operating within the time of a percept. Future progresses will depend on the careful use of comparative neuroscience (guiding the choices of experimental models and species adapted to the questions under study), on the definition of agreed-upon benchmarks for sensory stimulation, on the simultaneous acquisition of neural data at multiple spatio-temporal scales, and on the in vivo identification of key generic integration and plasticity algorithms validated experimentally and in simulations.

Sensory recalibration integrates information from the immediate and the cumulative past

Vision usually provides the most accurate and reliable information about the location of objects in our environment, and thus serves as a reference for recalibrating auditory spatial maps. Recent studies have shown that recalibration does not require accumulated evidence of cross-modal mismatch to be triggered, but occurs as soon as after one single exposure. Here we tested whether instantaneous recalibration and recalibration based on accumulated evidence represent the same underlying learning mechanism or involve distinct neural systems. Participants had to localize two sounds, a low- and a high-frequency tone, which were paired with opposite directions of audiovisual spatial mismatch (leftward vs. rightward). In accordance with the cumulative stimulus history, localization in unimodal auditory trials was shifted in opposite directions for the two sound frequencies. On a trial-by-trial basis, however, frequency-specific recalibration was reduced when preceded by an audiovisual stimulus with a different sound frequency and direction of spatial mismatch. Thus, the immediate past invoked an instantaneous frequency-invariant recalibration, while the cumulative past invoked changes in frequency-specific spatial maps. These findings suggest that distinct recalibration mechanisms operating at different timescales jointly determine sound localization behavior.

Cortical hierarchies perform Bayesian causal inference in multisensory perception

To form a veridical percept of the environment, the brain needs to integrate sensory signals from a common source but segregate those from independent sources. Thus, perception inherently relies on solving the "causal inference problem." Behaviorally, humans solve this problem optimally as predicted by Bayesian Causal Inference; yet, the underlying neural mechanisms are unexplored. Combining psychophysics, Bayesian modeling, functional magnetic resonance imaging (fMRI), and multivariate decoding in an audiovisual spatial localization task, we demonstrate that Bayesian Causal Inference is performed by a hierarchy of multisensory processes in the human brain. At the bottom of the hierarchy, in auditory and visual areas, location is represented on the basis that the two signals are generated by independent sources (= segregation). At the next stage, in posterior intraparietal sulcus, location is estimated under the assumption that the two signals are from a common source (= forced fusion). Only at the top of the hierarchy, in anterior intraparietal sulcus, the uncertainty about the causal structure of the world is taken into account and sensory signals are combined as predicted by Bayesian Causal Inference. Characterizing the computational operations of signal interactions reveals the hierarchical nature of multisensory perception in human neocortex. It unravels how the brain accomplishes Bayesian Causal Inference, a statistical computation fundamental for perception and cognition. Our results demonstrate how the brain combines information in the face of uncertainty about the underlying causal structure of the world.