Emotional speech synchronizes brains across listeners and engages large-scale dynamic brain networks

Abstract

Speech provides a powerful means for sharing emotions. Here we implement novel intersubject phase synchronization and whole-brain dynamic connectivity measures to show that networks of brain areas become synchronized across participants who are listening to emotional episodes in spoken narratives. Twenty participants' hemodynamic brain activity was measured with functional magnetic resonance imaging (fMRI) while they listened to 45-s narratives describing unpleasant, neutral, and pleasant events spoken in neutral voice. After scanning, participants listened to the narratives again and rated continuously their feelings of pleasantness-unpleasantness (valence) and of arousal-calmness. Instantaneous intersubject phase synchronization (ISPS) measures were computed to derive both multi-subject voxel-wise similarity measures of hemodynamic activity and inter-area functional dynamic connectivity (seed-based phase synchronization, SBPS). Valence and arousal time series were subsequently used to predict the ISPS and SBPS time series. High arousal was associated with increased ISPS in the auditory cortices and in Broca's area, and negative valence was associated with enhanced ISPS in the thalamus, anterior cingulate, lateral prefrontal, and orbitofrontal cortices. Negative valence affected functional connectivity of fronto-parietal, limbic (insula, cingulum) and fronto-opercular circuitries, and positive arousal affected the connectivity of the striatum, amygdala, thalamus, cerebellum, and dorsal frontal cortex. Positive valence and negative arousal had markedly smaller effects. We propose that high arousal synchronizes the listeners' sound-processing and speech-comprehension networks, whereas negative valence synchronizes circuitries supporting emotional and self-referential processing.

Keywords: Connectivity; Emotion; Network; Speech comprehension; Synchronization.

Fig. 1

Experimental design for fMRI (top) and subjective emotional ratings (bottom). Participants listened to short spoken narratives describing pleasant, unpleasant and neutral events. The discourses were preceded by a 5-s presentation of a fixation cross and by a 15-s presentation of a text that described the general context of the upcoming discourse without revealing details of its actual events. After fMRI, the participants listened to the discourses again, and rated their moment-to-moment experiences of valence (pleasantness–unpleasantness) and arousal.

Fig. 2

Means and 95% confidence intervals for valence and arousal rating (1–9) while the subjects were listening to the narratives. Vertical lines denote breaks between stories. Note that time is given as multiples of TR (1 TR = 1.7 s).

Fig. 3

Brain regions showing statistically significant (p < 0.05, FDR corrected) group-level ISPSs during listening of the narratives.

Fig. 4

Brain regions where ISPS was statistically significantly associated with self-reported valence and arousal time series during listening to the stories (p < 0.05, FDR corrected). Hot colours show positive and cool colours negative associations. The yellow (positive) and white (negative) outlines show the respective associations in conventional BOLD-GLM. Note: SMA = supplementary motor area, and SSC = somatosensory cortex.

Fig. 5

Connectivity graphs showing brain regions whose functional interconnectivity increased as a function of negative (left) or positive (right) emotional valence (top row, a–b) and arousal (bottom row, c–d). Colour coding denotes link type, and node circle size indicates node strength. The data are thresholded at q < 0.1, FDR corrected.

Fig. 6

Reduced connectivity graphs showing the main network hubs whose interconnectivity increased as a function of negative (left) or positive (right) emotional valence (top row, a–b) and arousal (bottom row, c–d).

Fig. 7

Normalized node degree of the largest hubs of the positive and negative valence and arousal networks. Note: Only twenty-five nodes with highest total degree are shown.

Fig. 8

Node-degree maps highlighting how many connections from each brain voxel were statistically significantly (p < 0.05 FDR corrected) modulated by negative and positive valence (top) and arousal (bottom).

Fig. 9

Brain regions where BOLD signal amplitude depended linearly on valence (top) and arousal (bottom) dimensions of emotional feelings elicited by the stories.