Representation of emotional expressions across the face and voice brain networks

Recognizing emotions from faces and voices is fundamental to successful social interaction, yet how the brain represents emotional information across sensory systems remains unclear. Here, we systematically mapped the contribution of individually-defined face- and voice-selective cortical regions in 24 human participants of either sex to unimodal and crossmodal emotion expressions. Using multivariate pattern classification, we show that emotional expressions are robustly decodable from the dominant sensory modality across all individually defined face- and voice-selective areas, and that these representations cannot be accounted for by low- and mid-level stimulus properties alone. Strikingly, emotional information from the non-dominant modality—vocal expressions within face networks and facial expressions within voice networks—was also represented throughout the network, with the exception of the occipital face areas. Crossmodal decoding further revealed a shared neural code for facial and vocal emotions in the temporal voice areas, posterior superior temporal sulcus, and precentral gyrus, indicating that these regions contain emotion representations that generalize across sensory modalities. Furthermore, in most face- and voice-selective regions, audiovisual emotional expressions could be distinguished from representations elicited by the dominant unimodal modality, indicating that bimodal signals are not simply represented as their strongest sensory component. Together, these findings reveal that emotion representations are distributed broadly across face- and voice-selective regions and identify temporo-frontal regions as key sites supporting modality-independent emotion representations. Significance statement Recognizing emotions from faces and voices is essential for everyday social interaction, yet the brain mechanisms that integrate these signals remain poorly understood. Using functional MRI (fMRI) and advanced pattern analyses, we show that emotional expressions are represented widely across both face- and voice-selective brain regions-even when conveyed through a non-preferred sensory modality. Strikingly, specific temporo-frontal regions encode facial and vocal emotions using a partially shared, modality-independent neural code. We further demonstrate that seeing and hearing emotions together reshapes how emotional information is organized in the brain, revealing a graded transition from sensory-specific to multisensory representations. These findings challenge strictly segregated models of emotion processing and provide a more integrated account of how the brain supports social understanding.