Presently, despite a large number of studies, the problem associated with the detection of speech patterns on the EEG is still relevant. In this study, we aimed to detect the spatial-frequency localization of specific EEG patterns associated with cognitive load related to the process of real and inner speech. To achieve this, we developed a methodology to identify epochs of stimulus-independent signals containing these patterns. Frequency localization was determined using the Ochiai coefficient, and spatial localization using one-way analysis of variance. The specific EEG pattern represent the frequency coherence coefficients in the resulting spectral-spatial window. It was shown that the realization of verbal activity causes the formation of spatiotemporal patterns of EEG activity in the gamma rhythm sub-band of frequencies 60–65 Hz in bilateral channels, which correspond to the following cortical regions: frontal superior (Fp1, Fp2), temporal superior (F7, F8, Ft7, Ft8), mid-temporal (T3, T4), temporo-parietal (Tp7, Tp8), temporal inferior (T5, T6), temporo-central (Cp3, Cp4), parietal (P3, P4, P5, P6), parieto-occipital (Po3, Po4, Po7, Po8). Mapping and optimization of the feature space for EEG patterns associated with inner speech was performed. Stable and efficient registration of such patterns makes it possible to solve the problem of establishing a neural control channel for brain-computer neural interface systems.