Figure 2. Cortical activations from intracranial recordings during speech production.

a. Speech neural activations in the high-gamma band (HG: 70 – 150 Hz; z-score normalized with respect to baseline) of an example electrode in the pre-central gyrus that showed significant modulation during the speech utterance at the single trial level (FDR-corrected non-parametric permutation test). b. Location of all 3534 intracranial electrodes in both left (1749) and right (1785) hemisphere, with significant HG modulation during the speech utterance. The electrodes are color-coded with respect to anatomical regions-of-interest (ROIs), and are assigned to planning, articulation, and monitoring networks based on previous literature. The numbers in the brackets indicate the number of significant electrodes with respect to the total number of recorded electrodes per ROI. c. Top: Mean HG activations of ROIs during response onset in left hemisphere, averaged within patient. Shaded area denotes standard error. Colored horizontal bars indicate time-periods of significant activation above baseline (one-sided permutation test with cluster correction, cluster forming threshold p<0.05, 1000 permutations). Bottom: Electrode distributions of peak HG time-points per ROI in left hemisphere. Significant differences in peak activations were observed and there was a general trend of temporal progression of neural activations across ROIs around the speech utterance (one-way ANOVAs: F_9,717 = 10.87, p < 0.001, two-sided). d. Electrode distributions of peak HG time-points per ROI groups. This temporal progression was also observed via significant differences in peak activations (one-way ANOVAs: F_2,724 = 30.7, p < 0.001, two-sided). Post-hoc one-sided t-tests indicate pairwise significance between the speech networks in left hemisphere: T_Planning < T_{Articulation,} t₍₃₆₅₎ = 2.938, p = 0.002; T_Articulation < T_Monitoring, t₍₅₀₇₎ = 9.25, p < 0.001, T_Planning < T_Monitoring, t₍₃₉₄₎ = 9.84, p < 0.001). e. Functional connectivity analysis validates the temporal progression of speech networks: peak lag of the net cross-correlation of the HG signals was computed within each subject. Dots represent individual subject median peak lags; box plots show the subject distribution of lags for each network pair comparison. The box spans the interquartile range (IQR), and whiskers extend to the observed value within 1.5 × IQR of each quartile boundary. Negative lags indicate that the first network (e.g. planning) leads the second (e.g. articulation). In all three pairwise comparisons, real peak lags were significantly earlier than the control obtained under a circular-shift null (one-sided paired Wilcoxon signed-rank test): Planning-Articulation (n=19 subjects, median = −180ms, W = 36, p = 0.007); Articulation-Monitoring (n=16, median = −398 ms, W = 16, p = 0.003); and Planning-Monitoring (n=15, median = −370ms, W = 30, p = 0.047). Together, these analyses demonstrate the temporal progression of neural responses for planning, articulation, and monitoring for speech production in the left hemispheric speech network.