Spontaneous recruitment of the person-knowledge network by familiar voices in the congenitally blind

Abstract

Recognising a familiar individual involves the spontaneous retrieval of person-specific information, engaging a distributed cortical network for person knowledge. While the role of this system is well established for visual face perception, it remains unclear whether its recruitment depends on visual experience. Here, we investigated the neural basis of voice-identity processing in congenitally blind and sighted individuals using fMRI. Participants listened to personally familiar and unfamiliar voices while performing an identity recognition task. Familiar voices elicited stronger responses across groups in the temporoparietal junction (TPJ), medial prefrontal cortex (mPFC), and right middle temporal gyrus. Of these, the TPJ and anterior mPFC—regions typically implicated in the network for person knowledge—showed a more pronounced familiarity effect in congenitally blind participants than in sighted controls. Multivariate analysis further revealed stronger identity-specific representations in the blind in the right TPJ, adjacent to the region showing enhanced univariate responses, as well as in the left intraparietal sulcus. These findings demonstrate that voice-identity cues alone can drive spontaneous access to person knowledge in the absence of visual input in both the normally sighted and the blind, but engagement of this network is stronger in the blind. These results indicate that the network for person knowledge develops in a modality-flexible manner shaped by the social demands of interpersonal interaction.

Humans are highly social animals. Our ability to identify and understand others is fundamental to social interaction. Recognising an individual is only the first step; when we see a familiar face, we spontaneously retrieve a wealth of associated information—from their personal history to our emotional attitudes towards them. This ability to access person-specific knowledge plays a key role in guiding social behaviour and communication. Neuroimaging research has identified a distributed network of brain regions involved in the spontaneous retrieval of such information^1–8. However, it remains unclear whether spontaneous activation of the neural systems for person knowledge is driven specifically by visual face perception, is developmentally contingent on input from the visual face-processing system, or instead reflects a general mechanism that can emerge independently across sensory modalities.

The cortical system for face and person perception has been broadly divided into a core system, responsible for visual analysis, and an extended system, implicated in the retrieval of non-visual, person-related information^1,7–10. The core system encompasses regions more closely associated with visual perception, the occipital and fusiform face areas (OFA/FFA) and the posterior superior temporal sulcus (pSTS). While these regions are consistently engaged during face perception, they do not reliably show stronger responses to familiar than unfamiliar faces. In particular, the FFA shows mixed results across studies, with some reporting increased, decreased, or unchanged responses to familiar stimuli^1,4.

In contrast, some regions of the extended system—including the temporoparietal junction (TPJ; comprising aspects of the angular gyrus and posterior STS), medial prefrontal cortex (mPFC), posterior cingulate cortex/precuneus, and anterior temporal lobes (ATL)—consistently show enhanced activity when participants view familiar faces^2–4,11. These regions represent key hubs of the default mode network (DMN), a set of areas implicated in internally directed cognition, including autobiographical memory, theory of mind, and semantic processing^12–15. Their recruitment when viewing familiar faces is thought to reflect the spontaneous retrieval of social, episodic, and biographical information about the recognised individual, supporting the notion of a distributed cortical network for person knowledge within the extended system. Studies cuing access to different forms of person-related knowledge show similar cognitive profiles for famous faces¹⁶ and famous names¹⁷, with the recruitment of these regions triggered by access to nominal, semantic, episodic, social and perceptual forms of person-knowledge with subtle regional biases (e.g., social knowledge in ATL, Precuneus, and mPFC; episodic information in TPJ). Strikingly, multivariate identity representations in this network, aligned across participants who know the same individuals, suggest that shared person-specific knowledge gives rise to common neural codes within this network⁸.

Faces are not the only means by which humans recognise each other. We are also able to reliably recognise a large number of individuals based on their voice alone. Processing of human voices, particularly speech, leads to increased activation in the posterior, middle, and anterior superior temporal gyrus and sulcus (STG/S), encompassing a region known as the temporal voice area (TVA). This region responds more strongly to human voices than to other auditory categories such as non-vocal sounds, scrambled voices, or animal cries¹⁸. The TVA, along with primary auditory cortex, has been proposed to form a core voice-perception system¹⁹. Responsivity of the right TVA increases during tasks that require recognition of speaker identity rather than attention to linguistic or acoustic features^20–24, and multivariate patterns of activation in TVA can distinguish between the voices of different individuals in both hemispheres^25–27, suggesting that voice identity may be represented within TVA²⁴. However, as with the core face-perception system, findings of increased univariate response to familiar voices—presumed to reflect stronger identity representations—have been mixed. Greater responses to unfamiliar voices have been reported in right TVA²¹, while other studies have reported greater responses to familiar voices^22,28,29, or no significant differences^30,31. Presently, it remains unclear whether identity-related processing in TVA reflects the acoustic analysis necessary for voice discrimination, or the representation of the individual’s identity itself³².

One possible reason for the disparate results from studies investigating voice familiarity is the modest sample sizes in some studies. Focussing on the two studies with larger samples sizes (N > 25) indicates reliably greater responses for familiar than unfamiliar voices in core TVA regions. Bethmann and colleagues²⁸ found widespread increases in response to familiar voices within posterior STG in the right hemisphere, bilaterally in the middle and middle-posterior STG, as well as bilaterally in the anterior STG—a region anatomically consistent with the ATL of the extended system that has been proposed to represent the identities of others^{23,28,32–34}. This study did not report whole-brain differences or effects elsewhere within the network for person knowledge. ROI analysis in the second study²⁹ found significantly increased responses for familiar voices again in right anterior STG/STS but also in a mid/posterior STG and only reported whole brain familiarity effects in the right supramarginal gyrus, a region not previously associated with the person-knowledge network. Collectively, these well-powered studies indicate an increased response for familiar voices across TVA. While these studies report modulation in anterior STG consistent with the ATL of the extended system, the shared response profile with neighbouring voice-sensitive TVA leaves it unclear whether this reflects voice processing or incidental access to person knowledge.

An alternate approach to isolating higher-level representations of person identity from voice perception is to examine crossmodal representations that generalise across sensory modalities, such as face and voice. Such amodal representations of famous identities have been found in the in the right pSTS³⁵, adjacent to but posterior to classic TVA. A study using personally familiar individuals³⁶ reported overlapping voice and face identity-decoding for voices and faces significantly in at least four out of the five studied participants in right pSTS, as well as observing crossmodal face-voice identity representations in sections of the left and right middle and anterior STS. This provides some insight into a potential non-perceptual representation of individual voice identity, yet the connection between the core-voice system and the extended system remains uncertain.

It has been proposed that the core-voice system is connected to the extended system³². Outside the TVA, increased responses to familiar compared to unfamiliar voices have occasionally been observed in extended system regions, such as the precuneus^21,22, although other studies have failed to replicate these effects^28,31. Associating biographical information with previously unfamiliar voices enhances activity in voice-identity–sensitive regions of the anterior STS and in speech-sensitive areas of the STG, but does not reliably engage regions of the extended system²⁹. In a recent study, Kanber and colleagues³⁷ compared responses to personally familiar voices versus newly learned or novel ones and reported increased within- and between-speaker dissimilarity for familiar voices in the precuneus, mPFC, and TPJ. These results could reflect either more fine-grained voice representations with increasing familiarity or, conversely, weaker familiar-identity representations that are more susceptible to noise. Collectively, these studies provide some, but not definitive, evidence that the extended system is spontaneously recruited during the identity processing of familiar voices.

Given that voice-based recognition is highly accurate, it is perhaps surprising that while recognition via faces reliably evokes the spontaneous retrieval of person knowledge, the evidence of voice-induced recruitment of the person-knowledge network is sparse. This asymmetry potentially reflects the privileged role of visual faces in perception and human social interaction. One possibility is that the primacy of faces leads to stronger neurodevelopmental connections between core-face perception systems and the extended system, and that spontaneous access to person-knowledge is driven by intrinsic connectivity between these systems. Alternatively, differences between face- and voice-linked retrieval of person knowledge may reflect their behavioural relevance to social interaction—seeing a person’s face may immediately prompt the retrieval of associated information helpful in social interaction, while voice recognition, as a secondary route for person identification, may not. One way to dissociate these possibilities is to examine individuals without visual experience. If congenitally blind individuals show spontaneous recruitment of the person knowledge network in response to familiar voices, this would indicate that visual experience is not required for the emergence of this network. Furthermore, if activity within the person knowledge network is greater in the blind than the sighted, this could reflect either enhanced behavioural reliance on voice for social interaction or experience-dependent neurodevelopmental plasticity that supports voice-based access to person knowledge.

In the present study, we investigate recruitment of the extended system for person knowledge in the congenitally blind. Blind and sighted participants performed an identity-repetition detection task on phrases spoken by personally familiar individuals and strangers while undergoing fMRI scanning. We predict that familiar voices will spontaneously recruit the extended system in blind individuals, indicating that such access to person knowledge does not depend on vision or neurodevelopmental connections to the face-processing system. We further predict that this response will be greater in the blind than the sighted, consistent with either experience-driven plasticity or the increased behavioural relevance of voice-based recognition in the absence of vision.

Methods

Participants

Seven blind (age: x̅ = 35.9, sd = 9.5 years; five female) and seven sighted (age: x̅ = 35.0, sd = 13.1, four female) participants were recruited in this study. Six blind subjects were so from birth, one lost sight at 16 months of age. Three had residual light perception in one or both eyes, but none could perceive form or motion. All participants had no history of neurological or psychiatric disorders. Sighted participants had normal or corrected-to-normal vision. All participants gave informed consent and were reimbursed for their time. The study was approved by the Ethics Committee of the University of Trento, and all research was conducted in accordance with relevant guidelines and regulations, including the Declaration of Helsinki (1964, as revised in 2013).

Stimuli

Stimuli consisted of 64 Italian sentences spoken by four individuals highly familiar to the participant and four unfamiliar speakers (mean sentence duration = 1.61 s, SD = 0.36). For each speaker, four sentences were socially engaging (e.g., “Hi [participant’s name], how are you?”), and four were neutral (e.g., “Sardinia is an island”). Recordings were made using GarageBand (Apple Inc.) and segmented into individual sentences using Audacity (Audacity Team). The familiar voices of each participant were used as unfamiliar voices for other participants, ensuring that stimuli did not systematically differ across familiar and unknown conditions. Stimuli were balanced across participants for age, gender, and sentence type.

Procedure

Before entering the scanner, participants performed an identity recognition task. In this task, participants indicated with button press whether they recognized the identity of the speaker for each sentence that would be used in the main experiment. Once participants reached a recognition accuracy of at least 90%, they were familiarized with the experimental task to ensure understanding of task instruction, then entered the scanner for the experiment.

The experiment task consisted of 1-back repetition detection of the identity of the speaker for both familiar and unfamiliar identities. Trials were six seconds in duration and consisted of the spoken phrase followed by a period of silence. On each trial, participants responded with a button-press indicating whether or not the identity was repeated (buttons counterbalanced across subjects). The experiment was divided in four runs of 12 min, 40 s. Runs consisted of 80 stimulus presentations (10 from each of the eight speakers) and 26 null events consisting of 6 s of silence. The inclusion of null events resulted in intervals between the onset of one stimulus and the next ranging from 6 to 18 s. The sentences were presented in counterbalanced order among familiar and unknown voices, engaging, neutral and blank sentences.

MRI scanning

Stimuli were played through a SereneSound Audio System (Resonance Technology, Inc., California, USA.) at a level fixed across participants so as to be clearly audible above the scanner noise. Data collection was conducted at the Center for Mind/Brain Sciences (CIMeC), University of Trento on a Bruker BioSpin MedSpec 4T using a USA Instruments 8-channel phased-array head coil. One thousand two hundred and sixty volumes of 37 AC–PC aligned slices were acquired over 4 runs using an echo-planar 2D imaging sequence (image matrix = 70 × 64, repetition time = 2200 ms, echo time = 33 ms, flip angle = 76°, slice thickness = 3 mm, gap = 0.45 mm, with 3 × 3 mm in plane resolution). An additional high-resolution (1 × 1 × 1 mm³) T1-weighted MPRAGE sequence was acquired (sagittal slice orientation, centric phase encoding, image matrix = 256 × 224 [Read × Phase], field of view = 256 × 224 mm [Read × Phase], 176 slices with 1-mm thickness, GRAPPA acquisition with acceleration factor = 2, duration = 5.36 min, repetition time = 2700, echo time = 4.18, TI = 1020 ms, 7° flip angle).

fMRI analysis

Preprocessing

Analysis was performed in SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12/). After discarding the first 4 volumes of each run to allow for T1 equilibration, all images were corrected for head movement. Slice-acquisition delays were corrected using the middle slice as reference. All images were normalised to the standard SPM8 EPI template (Montreal Neurological Institute [MNI] stereotactic space), resampled to a 3-mm isotropic voxel size, and spatially smoothed using an isotropic Gaussian kernel of 8-mm full-width half-maximum (FWHM) for the univariate analysis and 4 mm for the multivariate analysis. The time series at each voxel for each participant were high pass filtered at 128 s and prewhitened by means of an autoregressive model AR(1).

Univariate analysis

Subject-specific beta weights were derived with a general linear model (GLM). Eight regressors were formed by convolving the onset of each of the eight speaker-identities with the SPM12 hemodynamic response function. The six head-motion parameters were included as additional regressors of no interest. For inferential analysis of the familiarity effect and the group-by-familiarity interaction, beta values were averaged for the four familiar voices and for the four unfamiliar voices. These condition-wise averages were then entered into a mixed within-between group-level random-effects GLM. For between-subject group analysis, beta values were averaged across all conditions and entered into a two-sample t-test random effects model.

Multivariate pattern analysis

MVPA^38–41 was performed for single trials on the pre-processed EPI volumes (10 images per speaker per run). To account for the haemodynamic lag, the third EPI volume after stimulus onset was chosen. Linear discriminant analysis (LDA⁴²) was performed separately for four familiar voices and for the four unfamiliar voices. Leave-one-run-out cross-validation was performed across the four runs. A searchlight analysis⁴³ was used to identity voxels showing above chance decoding accuracy. Output decoding-accuracy searchlight maps were smoothed with a 4-mm FWHM kernel, masked using a custom grey-matter, and entered into random effects analysis. Between-sample main and simple effects of group were identified using independent t-tests on decoding-accuracies averaged across familiar and unfamiliar identities. Within-subject effects of familiarity were investigated via mixed within-between group-level random-effects GLM.

An additional control analysis to account for the potential effect of a greater number of recognised trials in blind than sighted participants was performed. Here, we repeated the above analysis considering only correct trials, and matched the number of trials between groups by randomly removing 10% (8 per run) of the trials from blind participants from the analysis. Additionally, to account for potential effects of the uneven number of trials in training and testing chucks, for each cross-validation iteration the number of samples was balanced across training/testing partitions by randomly leaving out the excessive samples, with this process being repeated until all samples had been included at least once.

Region of interest analysis

Orthogonal contrasts (main effects of group and familiarity), were used to define Regions of Interest (ROI)s. ROIs were formed from the union of a 5-mm sphere centred on peak coordinates and the activation map of the same contrast (thresholded at p < 0.001), using MarsBar (https://marsbar-toolbox.github.io). Beta values were then extracted, averaged across the ROI, and then assessed using Student’s t tests or ANOVA.

Results

Behavioural

Identity repetition detection accuracy was high in both blind (95.0%) and sighted participants (86.6%) and did not differ between groups (p = 0.11). The average response time was 2.13 s (SD = 0.26) from the stimulus onset and did not differ as a function of familiarity or group (all t values < 1).

Main effect of group

First, we addressed the differences in cortical activity between groups. Greater activation was observed in the visual cortices of blind participants, and was more pronounced in the right hemisphere (Fig. 1, Table 1). Peaks were located along the medial surfaces in the cuneus, a region of the lingual gyrus abutting the calcarine sulcus and the lateral fusiform gyrus. An additional peak was evident in the lateral occipital cortex in the middle occipital gyrus. These findings are consistent with the well-characterised plasticity of visual cortex in blind individuals⁴⁴. No regions showed significantly greater activation in sighted compared to blind participants.

Fig. 1.

Main Effect of Group. Regions showing greater activation in blind compared to sighted participants. Initial voxel-wise threshold p < .001, extent 200, cluster corrected for multiple comparisons (p < .05, FWE). Axial cross section taken at x = − 25.

Table 1.

Main effect of group. Left and right hemispheric peak locations within the occipital cluster showing greater activation in blind compared to sighted participants (cluster extent: 5995 voxels, p < .00001 corrected).

Region	Left					Right
	T	p(unc)	x,y,z	{mm}		T	p(unc)	x,y,z	{mm}
Lateral occipital	12.88	< .0001	− 39	− 82	4	12.07	< .0001	39	− 46	− 26
Cuneus	9.22	< .0001	− 21	− 82	25	12.27	< .0001	27	− 85	28
lat. fusiform gyrus	10.02	< .0001	− 33	− 58	− 17	12.20	< .0001	36	− 46	− 26
Lingual gyrus	8.17	< .0001	− 12	− 73	7	10.58	< .0001	6	− 82	− 2

We performed an ROI analysis (see methods) comparing familiarity effects within the eight peaks reported in Table 1. Neither group-by-familiarity interactions (all p values > 0.1) nor simple effects of familiarity in the blind (all p values > 0.25) were present. This suggests a more task-general activation during speech processing within these regions, rather than a specific role in familiar identity processing.

Main effect of familiarity and group by familiarity interaction

The main effect of familiarity (familiar > unknown) is presented in Fig. 2A and Table 2. Personal familiarity with the speaker increased activity in posterior and anterior clusters in the mPFC, bilateral TPJ, and a contiguous cluster in the right temporal lobe with peaks in the posterior and mid middle temporal gyrus (MTG). ROI analysis was performed to assess differential effects of familiarity for sighted and blind groups. Critically, this interaction is orthogonal to the contrast used to define the ROI and is therefore free from circularity (Friston et al. 2006). Pronounced interactions were seen in the left (F_(1,12) = 41.9, p < 0.0001) and right (F_(1,12) = 13.0, p = 0.0036) TPJ, as well as the anterior cluster in the mPFC (F_(1,12) = 11.92, p = 0.0048). Signal plots (Fig. 2B) reveal that this interaction reflects a stronger response to familiar compared to unfamiliar voices in blind individuals relative to sighted. Figure 2C shows the distribution of the familiarity effect in blind and sighted participants.

Fig. 2.

Brain regions modulated by familiarity (main effect) and differential familiarity effects by group within these regions (interaction). (A) Univariate contrast of familiar > unknown. Initial voxel-wise threshold p < .001, cluster corrected for multiple comparisons (p < .05, FWE). (B) Signal plots showing familiarity by group interactional. Error bars indicate within-subject standard errors and are descriptive only due to ROI definition-interrogation circularity. Only the interactions (grey significance brackets and familiarity-effect differences in panel C) are valid for inference. (C) Box and whisker plot showing distribution of familiarity effect (familiar minus unknown) for participants within each group.

Table 2.

Main effect of familiarity. Location, extent and significance for the contrast: familiar > unknown.

Region	Cluster		Peak
	p(FWE-corr)	voxels	T	p(unc)	x, y, z (mm)
Left TPJ	.002	123	8.87	< .0001	− 54	− 64	16
Right pMTG	< .001	178	7.76	< .0001	60	− 40	− 2
Right mSTG			6.32	< .0001	51	− 16	− 17
Right TPJ	.010	89	6.84	< .0001	63	− 55	25
amPFC	.001	137	6.77	< .0001	− 9	65	13
pmPFC	.013	84	6.51	< .0001	− 12	38	4

Whole-brain analysis indicated that no regions showed significantly greater activation for unfamiliar than familiar voices.

Identity decoding in blind and sighted

To identify cortical locations where patterns of neural activity distinguished speaker identity, MVPA was performed separately for the four familiar and the four unknown speakers. Figure 3 shows above-chance identity decoding accuracy irrespective of familiarity. Both sighted and blind participants showed significant identity decoding in bilateral temporal lobes. Descriptively, this appeared more widespread in blind participants, extending into TPJ, mPFC, the precuneus as well as the left intraparietal sulcus and occipital pole (compare Fig. 3A and B). Significant group differences, however, were found only in the right angular gyral component of the TPJ and left lateral intraparietal sulcus (LIP; Fig. 3C; Table 3). ROI analysis within the right angular gyrus and LIP did not indicate modulation by familiarity or group-by-familiarity interactions (all p values > 0.1).

Fig. 3.

Identity decoding accuracy, averaged across familiar- and unknown-identity decoding. (A) Identity decoding accuracy for sighted participants. (B) Identity decoding accuracy for blind participants. (C) Main effect of group: regions showing stronger identity representation in blind than sighted participants. Initial voxel-wise threshold p < .001, cluster corrected for multiple comparisons (p < .05, FWE).

Table 3.

Identity decoding accuracy—main effect of group. Location, extent and significance for the contrast: blind > sighted.

Region	Cluster		Peak
	p(FWE-corr)	Voxels	T	p(unc)	x, y, z (mm)
Right AG	.016	64	7.50	< .0001	54	− 64	37
Left LIP	< .001	209	7.38	< .0001	− 30	− 70	43

To determine whether similar group differences in identity decoding accuracy were present in the contralateral hemisphere that were below the whole-brain statistical thresholds, AG and IPS ROIs were left–right flipped and decoding accuracy assessed via sample t-test. While effects were absent in the right hemisphere homologue of the left IPS region (t₍₁₂₎ = 0.55, p = 0.59), decoding accuracy in the left hemisphere AG approached significance (t₍₁₂₎ = 1.91, p = 0.08).). In this way, stronger identity representations within this right hemisphere AG homologue in blind than sighted individuals cannot be fully excluded.

Although we did not observe significant differences in task-performance between blind and sighted participants, blind participants had nominally greater accuracy and group differences in decoding accuracy may reflect the greater number of recognised trials in blind participants. To address this, we repeated the MVPA analysis removing incorrect trials (and balancing the number of trials across groups, see methods). This analysis revealed that, when only correct trials were considered, significant group differences persisted in right AG (p_FWE = 0.023; extent = 51 voxels; peak_xyz: 54, − 64, 37) and left IPS (p_FWE < 0.001; extent = 172 voxels; peak_xyz: − 33, − 73, 37), indicating that the above effects cannot be fully accounted for by increased accuracy in the blind group.

Whole-brain analysis of the main effect of familiarity and group-by-familiarity interaction revealed no significant effects.

Discussion

We investigated whether voice-identity processing of familiar individuals leads to the spontaneous recruitment of the distributed cortical network for person knowledge in sighted and congenitally blind participants. We compared the fMRI response to familiar and unfamiliar voices while participants performed a one-back identity recognition task. Analysis showed that hearing personally familiar voices evoked stronger responses in areas of the middle temporal gyrus bordering the TVA, bilateral TPJ, and mPFC. Of these regions, robust familiarity-contingent enhancement in blind relative to sighted individuals was observed in the TPJ and an anterior portion of the mPFC. Further, multivariate analysis revealed stronger identity representations in the blind than sighted participants in the right TPJ and left intraparietal sulcus.

The disparity in familiarity effects between blind and sighted participants was greatest in bilateral TPJ. Person-sensitive TPJ can be differentiated into: classic face-selective pSTS, which responds selectively to face-images¹⁰ and biological motion^45,46 but does not show modulation when accessing person-knowledge is cued¹⁶; and more posterior elements including the angular gyrus that are modulated by access to diverse types of person-related knowledge irrespective of whether familiar faces and names are presented^16,17,45. Familiarity effects in the present study were located in this latter region, an area associated with theory of mind¹⁵, accessing person-related concepts (e.g., ‘professions’⁴⁷), stored encyclopaedic knowledge relating to people⁴⁸, and viewing the personally familiar faces of others^2,3. While these prior findings are typically based on group-level analyses, and it remains unclear to what extent the apparent overlap reflects diverse cognitive subdivisions within the TPJ blurred by group-level averaging, the greater activation for familiar individuals observed in the blind broadly across this region is consistent with a preferential role of this region in spontaneous access to person-related knowledge when recognising the identity of known individuals.

In the right hemisphere, the TPJ region sensitive to familiarity in the blind was seen to abut, but not overlap with, the cortical area showing strong multivariate representations of speaker identity in the blind. This potentially indicates that familiarity processes in right TPJ are linked to identity representations in neighbouring regions of the angular gyrus. However, it is noteworthy that identity representations in this region were not seen to differ between familiar and unknown individuals, suggesting general processing and differentiation of others in this region and a complex relationship between familiarity and identity processing in the right TPJ of the blind. In the left hemisphere, stronger multivariate representation of identity was also seen. However, this region was not contiguous with the area of the left TPJ showing enhanced familiarity effects in the blind. Peak decoding accuracy was located in left LIP rather than the angular gyrus, leaving the relationship between familiarity enhancement and identity representation in the left hemisphere unclear.

Familiarity effects were also observed in mPFC, and were enhanced in the blind with respect to sighted participants in an anterior cluster. This region of mPFC, like the TPJ, is recruited by explicit access to a range of person-related knowledge, but is particularly driven by social cognition^16,17,45. It is consistently implicated in the representation of personal traits⁴⁹ and mental states of others^50,51, supporting its central role in mentalising and person understanding. This region is functionally connected to the default mode network at rest, with strong functional connectivity with the TPJ, as well as precuneus and anterior temporal lobes⁵². The enhanced response in the blind when hearing familiar people may reflect increased engagement of internally directed socially relevant representations typically supported by this network. While other person-related processes also occur within the region, its involvement in the blind might predominantly reflect access to social knowledge about familiar individuals, supporting its role in person-specific inference even in the absence of visual input.

These findings of increased responses for familiar individuals in the blind in TPJ and mPFC show that spontaneous recruitment of the system for person-knowledge is not constrained by the intrinsic connectivity between face perceptual systems and the extended system. In sighted individuals, the perception and further processing of faces and voices share overlapping demands and likely engage overlapping neural mechanisms^34,53. Both show highly selective, right-lateralised responses in their core processing systems. In the environment, faces and voices are often perceived together and crossmodally face-voice integration occurs obligatorily, as evidenced by phenomena like the McGurk effect (where the perception of consonant sounds is shaped by perceived lip movement⁵⁴) and by findings that visually attending to a congruent (versus incongruent) lip movement modulating the auditory response in TVA and the visual response in early visual cortex and the fusiform gyrus⁵⁵. The pSTS is driven both by voices and faces and has been implicated in face-voice integration³⁴ and functional connectivity between FFA and right pSTS may drive familiar-voice effects in the region²².

During face-perception, FFA drives activation in the extended network⁵⁶, and this connectivity may underlie the development of functional specialisation in the extended system necessary for the spontaneous retrieval of person knowledge. However, in the blind, while some recruitment of face selective regions like the FFA have been reported for person-related stimuli^57,58, these effects are often mixed and greatly reduced compared to the sighted (see⁵⁹, for a review). Rather, regions anatomically consistent with FFA respond selectively to speech over linguistic human vocalisations²⁷ and respond not only selectively during language tasks but do so as a function of the syntactic complexity of the spoken sentence^44,60. The present finding shows that spontaneous access to person knowledge is not contingent on visual experience and likely develops independently of strong input from the core face-perception network. This interpretation gains support from findings of enhanced responsivity in the person-knowledge system in the congenitally blind for words describing famous people compared to words describing face parts (but not when comparing famous buildings to everyday scenes)⁶¹.

The ventral temporal cortical areas that were more activated by voice perception in the blind, as compared to the sighted, are generally considered visual areas in the sighted, with fields specialised for perception of visual categories, such as faces, bodies, scenes, and objects. These results suggest that these cortices may participate in nonvisual functions in the blind. In previous work, we showed that fusiform cortex in blind participants also represents the emotion conveyed by voices²⁷. We and others have shown that fusiform cortex in sighted participants is engaged by processing the agentic properties of dynamic stimuli, independent of their role in face, object, and scene perception^45,62–66. These studies, combined with the results in this paper, support the hypothesis that ventral temporal cortex can play a more general role in social cognition independent of input modality⁵⁸.

The presence of robust differences in voice-identity driven recruitment of the system for person knowledge in the blind might result from two processes. One may reflect plasticity in neurodevelopment, where the absence of strong connections between core face- and person-knowledge–related networks may lead to stronger connectivity between core voice-perception regions and the person-knowledge network. This, in turn, facilitates the spontaneous retrieval of person related knowledge in TPJ and ventral mPFC when familiar voices are heard. Alternatively, this may not reflect underlying neuroarchitecture but instead the relative importance of voice identity processing for blind relative to sighted individuals. It is possible that sighted individuals rely on faces to cue preparation for social interaction (and person-knowledge retrieval) and voice-based identity detection in the sighted, while highly accurate, is a secondary process not used as frequently in day-to-day interaction that does not prompt the same social-interaction mechanisms. In contrast, this process is performed exclusively by voices in the blind who rely on voice identity detection to guide all their social interactions, and thus spontaneously retrieve the information necessary to facilitate social interaction following voice-identity extraction. In this way, face dominance in the sighted may relegate voice-based person knowledge retrieval to a secondary and less well-developed role. These alternatives remain an open question in the present study. Future studies investigating individuals who become blind later in life would help differentiate neurodevelopmental and behavioural-relevance explanations for enhanced person-knowledge activation in the blind.

We also observed a greater response to familiar people, that did not vary as a function of group, in middle and posterior MTG. It is notable that this region overlapped closely with the right temporal areas where identity decoding converged between sighted and blind participants (compare Fig. 3A and B). This indicates a convergence between regions commonly showing familiarity enhancement and identity representation for both individuals with and without visual experience. This region is not associated with the network for person knowledge, with effects seen inferiorly to classic TVA and it is possible that the locus of this effect, within MTG, reflects a gradient of voice abstraction, from early acoustic processing in Heschl’s Gyrus and the Planum Temporale, to higher level voice perception in the STS/G, to more abstract representations that support recognition of individuals based on stored voice representations across varied speech utterances.

In this work, we propose supporting evidence that, when listening to familiar voices, person knowledge is incidentally retrieved in the blind to a greater extent than sighted individuals. However, future work is needed to confirm that voice-cued spontaneous access to person knowledge is uncommon in the sighted, as the existing literature on this question is inconclusive. Specifically, studies showing, within the same participants, stronger recruitment of the network for person knowledge for face than voice stimuli for familiar people could provide evidence for such face dominance in the sighted. Additional studies showing behavioural advantages for face- over voice-based access to person knowledge would help provide a direct link between incidental retrieval and functional advantage. Moreover, while performance differences did not reach significance, blind participants were on average 10% more accurate than sighted individuals. One possibility is that increased task demand in the sighted may discourage the incidental retrieval of associated person-related knowledge. Future studies could address this by systematically manipulating identity-recognition difficulty to determine whether recruitment of the extended person-knowledge network scales with task demands rather than modality dependence.

This study employed a modest sample size of fourteen participants, seven of which were blind. This can be expected to substantially increase the probability of Type II errors—the failure to detect true effects that would be discernible at greater statistical power. The reported familiarity effects and identity decoding accuracies were driven by the highly stable nature of the responses in blind participants. In this context, the failure to observe statistically significant familiarity effects in the precuneus or ATL must be interpreted with caution, and it remains an open question whether familiarity may influence activation in these regions. Similarly, the presence of multivariate voice-cued representations of identity in the sighted population cannot be excluded—it is possible that future studies with larger samples may reveal such effects in sighted individuals. While we can have good statistical confidence in the positive effects reported in this study, as while small sample sizes increase variability in the measure, this is directly accounted for in the standard error and the test statistic, it is important to consider that the literature-level false-positive rate with less well powered studies may still be elevated when combined with publication bias and selective reporting⁶⁷.

To place these results within the broader context of plasticity effects in the blind, it is worth considering two models that describe the principles that may guide plastic reorganisation in the absence of sensory input: the metamodal⁶⁸ and pluripotent⁴⁴ hypotheses. The metamodal hypothesis posits that cortical regions are biased towards both sensory inputs and cognitive computations, and that in the absence of the preferred sensory input route, the computational/perceptual processes are driven via input from other sensory modalities. Canonical examples of such organisational principles in the blind include the recruitment of the visual word form area by tactile Braille⁶⁹ reading and recruitment of visual-motion sensitive MT during auditory motion detection⁷⁰. More recently, the pluripotent hypothesis has extended this principle to account for plasticity effects that cannot easily be accounted for by metamodal computational biases. For instance, sensitivity to linguistic grammatical complexity in the visual ventral stream⁶⁰ or mathematical complexity within the posterior middle occipital gyrus^71‚72. The pluripotent hypothesis proposes that cortical function is not only shaped by biases towards particular computational processes but by long range connections with higher order brain regions that sculpt regional specialisation over development. Our findings sit at the intersection of these two accounts. That is, in the absence of vision, TPJ and mPFC, regions that are associated with person-knowledge, may develop stronger connections with voice-perception regions, enhancing their role in the incidental retrieval of person knowledge. Thus, our findings support both shared functional organisation and experience-dependent reorganisation.

In this study, we investigated whether the spontaneous recruitment of the distributed cortical network for person knowledge during voice recognition requires visual experience. We found that familiar voices elicited stronger responses in key regions of the network for person knowledge—the TPJ and mPFC—to a greater extent in the congenitally blind. Multivariate analysis further revealed stronger identity representations in the blind in the right TPJ, adjacent to the region showing enhanced univariate responses, as well as in the left intraparietal sulcus. These findings suggest that, in the absence of visual input, voice-identity cues alone are sufficient to drive spontaneous access to person knowledge, engaging elements of the same distributed network typically recruited by face recognition in the sighted. This supports the view that the network for person knowledge operates as a flexible, modality-independent network for representing socially and semantically rich information about others. Together, these results demonstrate that the role of the network for person knowledge in social interactions can develop without visual input, highlighting its role as a modality-flexible network shaped by the social demands rather than sensory input.

Author contributions

I.G. original conceptualisation. I.G. and K.B.P. designed the study. S.L.F., K.B.P., and C.B. conducted the research. S.L.F. performed the analyses and wrote the first draft of the manuscript. I.G. reviewed and edited the manuscript.

Funding

This work was supported in part by the European Research Council (ERC) Starting Grant awarded to S.L.F. under the European Union’s Horizon 2020 research and innovation programme (Grant No. 640594, CRASK—Cortical Representation of Abstract Semantic Knowledge).

Data availability

Data supporting the findings of this study are available on reasonable request. Requests should be directed to Scott Fairhall.

Declarations

Competing interests

The authors declare no competing interests.