Fig. 2: Brain responses and functional networks for natural and deepfake speaker identities.
From: Cortical-striatal brain network distinguishes deepfake from real speaker identity
a Neural activity patterns for contrasting [IDnat > IDdf]. White dashed line indicates the voice-sensitive regions evoked by the functional voice localizer scan. Second-level group t-maps, p < 0.005 corrected at the cluster level k > 47. b Beta estimates in right NAcc, right mid STG, and left mid STG for the conditions that were included in the interaction contrast ([IDnat > IDdf] > [SPEECHnat > SPEECHdf]). Plots show individual parameter estimates (n = 25) extracted from the maximum statistic for the contrasts shown in (a). Asterisks indicate significant effects of LMMs, *p < .001, We ran LMMs to specify the interaction effects. Circles indicate individual data and horizontal lines mean values. c Functional connectivity networks supporting natural identity matching (second-level group t-maps, gPPI, p < 0.005 corrected at the cluster level k > 47) from contrasting IDnat > baseline]; neural seeds as highlighted in (a). d Functional connectivity patterns with higher and lower connectivity for IDnat>IDdf and IDdf>IDnat (second-level group t-maps, gPPI, p < 0.005 corrected at the cluster level k > 47); neural seeds as highlighted in (a). Cd: caudate nucleus, FG: fusiform gyrus, HC: hippocampus, IFG, orb: inferior frontal gyrus, pars orbitalis, tri: pars triangularis, ITG: inferior temporal gyrus, L: left, LOC: lateral occipital cortex, MTG: middle temporal gyrus, NAcc: nucleus accumbens, OFG: occipital fusiform gyrus, PCG: posterior cingulate gyrus, PrCG: precentral gyrus, PT: planum temporale, R: right, SMC: supplementary motor cortex, SMG: supramarginal gyrus, STG: superior temporal gyrus, TP: temporal pole.
