38 more complex and involves neural networks of distinct cortical areas engaged in phonological or semantic analysis, speech 39 production and other processes15, 18. A few studies have already been conducted within the inner speech paradigm using 40 EEG19–21, ECoG15, functional Magnetic Resonance Imaging (fMRI) and positron emission tomography scan22–25. 41 Another paradigm related to the inner speech is the so-called “auditory imagery”26, 27. In this paradigm, instead of actively 42 producing the speech imagination, the subject passively listens to someone else’s speech. It has already been explored 43 using EEG19, 28, ECoG29, 30 and fMRI31, 32. Although this paradigm is not particularly useful for real BCI applications, it has 44 contributed to the understanding of neural processes associated with speech-related paradigms. 45 While publicly available datasets for imagined speech10, 33 and for motor imagery34–38 do exist, to the best of our knowledge 46 there is not a single publicly available EEG dataset for the inner speech paradigm. In order to improve the understanding of 47 inner speech and its applications in real BCIs systems, we have built a multi speech-related BCI dataset consisting of EEG 48 recordings from ten naive BCI users, performing four mental tasks in three different conditions: inner speech, pronounced 49 speech and visualized condition. The last two of them are explained in detail in the Section BCI Interaction Conditions. These 50 conditions allow us to explore whether inner speech activates similar mechanisms as pronounced speech or whether it is closer 51 to visualizing a spatial location or movement. Each participant performed between 475 and 570 trials in a single day recording, 52 obtaining a dataset with more than 9 hours of continuous EEG data recording, with over 5600 trials. 53

Methods 54 Participants 55 The experimental protocol was approved by the “Comité Asesor de Ética y Seguridad en el Trabajo Experimental” (CEySTE, 56 CCT-CONICET, Santa Fe, Argentina1). Ten healthy right-handed subjects, four females and six males with mean age ± std 57 = 34 ± 10 years, without any hearing or speech loss, nor any previous BCI experience, participated in the experiment and 58 gave their written informed consent. Each subject participated in an approximately two hours recording. In this work, the 59 participants are identified by aliases “sub-01” through “sub-10”. 60 Experimental Procedures 61 The study was conducted in an electrically shielded room. The participants were seated in a comfortable chair in front of a 62 computer screen where the visual cues were presented. In order to familiarize the participant with the experimental procedure 63 and the room environment, all steps of the experiment were explained, while the EEG headcap and the external electrodes were 64 placed. The setup process took approximately 45 minutes. Figure 1 shows the main experiment setup. 65 The stimulation protocol was designed using Psychtoolbox-339 running in MatLab40 and was executed on a computer, 66 referred to as PC1 in Figure 1. The protocol displayed the visual cues to the participants in the Graphic User Interface (GUI). 67 The screen’s background was light-grey coloured in order to prevent dazzling and eye fatigue. 68 Each subject participated in one single recording day comprising three consecutive sessions, as shown in Figure 2. A 69 self-selected break period between sessions, to prevent boredom and fatigue, was given (inter-session break). At the beginning 70 of each session, a fifteen seconds baseline was recorded where the participant was instructed to relax and stay as still as 71 possible. Within each session, five stimulation runs were presented. Those runs correspond to the different proposed conditions: 72 pronounced speech, inner speech and visualized condition (see Section BCI Interaction Conditions). At the beginning of each 73 run, the condition was announced in the computer screen for a period of 3 seconds. In all cases, the order of the runs was: one 74 pronounced speech, two inner speech and two visualized conditions. A one minute break between runs was given (inter-run 75 break). 76 The classes were specifically selected considering a natural BCI control application with the Spanish words: "arriba", 77 "abajo", "derecha", "izquierda" ( i.e."up", "down", "right", "left", respectively). The trial’s class (word) was randomly presented. 78 Each participant had 200 trials in both the first and the second sessions. Nevertheless, depending on the willingness and 79 tiredness, not all participants performed the same number of trials in the third session. 80 Figure 3 describes the composition of each trial, together with the relative and cumulative times. Each trial began at time 81 t = 0 s with a concentration interval of 0.5 s. The participant had been informed that a new visual cue would soon be presented. 82 A white circle appeared in the middle of the screen and the participant had been instructed to fix his/her gaze on it and not to 83 blink, until it disappeared at the end of the trial. At time t = 0.5 s the cue interval started. A white triangle pointing to either 84 right, left, up or down was presented. The pointing direction of the cue corresponded to each class. After 0.5 s, i.e. at t = 1 s, 85 the triangle disappeared from the screen, moment at which the action interval started. The participants were instructed to start 86 performing the indicated task right after the visual cues disappeared and the screen showed only the white circle. After 2.5 s of 87 action interval, i.e. at t = 3.5 s, the white circle turned blue, and the relax interval began. The participant had been previously 88 instructed to stop performing the activity at this moment, but not to blink until the blue circle disappears. At t = 4.5 s the blue 89 circle vanished, meaning that the trial has ended. A rest interval, with a variable duration of between 1.5 s and 2 s, was given 90 between trials. 91 To evaluate each participant’s attention, a concentration control was randomly added to the inner speech and the visualized 92 condition runs. The control task consisted of asking the participant, after some randomly selected trials, which was the direction 93 of the last class shown. The participant had to select the direction using the keyboard arrows. No time limit was given to reply 94 to these questions and the protocol continued after the participant pressed any of the four arrow keys. Visual feedback was 95 provided indicating whether the question was correctly or incorrectly answered. 96 Data Acquisition 97 Electroencephalography (EEG), Electrooculography (EOG) and Electromyography (EMG) data were acquired using a BioSemi 98 ActiveTwo high resolution biopotential measuring system2. For data acquisition, 128 active EEG channels and 8 external active 99 EOG/EMG channels with a 24 bits resolution and a sampling rate of 1024 Hz were used. BioSemi also provides standard EEG 100 head caps of different sizes with pre-fixed electrode positions 3. A cap of appropriate size was chosen for each participant 101 by measuring the head circumference with a measuring tape. Each EEG electrode was placed in the corresponding marked 102 position in the cap and the gap between the scalp and the electrodes was filled with a conductive SIGNAGEL®4 gel. 103 Signals in the EOG/EMG channels were recorded using a flat-type active electrode, filled with the same conductive gel and 104 taped with a disposable adhesive disk. External electrodes are referred from “EXG1” to “EXG8”. Electrodes EXG1 and EXG2 105 were both used as a no-neural activity reference channels, and were placed in the left and right lobe of each ear, respectively. 106 Electrodes EXG3 and EXG4 were located over the participant’s left and right temples, respectively, and were intended to 107 capture horizontal eye movement. Electrodes EXG5 and EXG6 aimed to capture vertical eye movement, mainly blinking 108 movements. Those electrodes were placed above and below the right eye, respectively. Finally, electrodes EXG7 and EXG8 109 were placed over the superior and inferior right orbicularis oris, respectively. Those electrodes were aimed to capture mouth 110 movement in the pronounced speech and to provide a way for controlling that no movement was made during the inner speech 111 and visualization condition runs. 112 The software used for recording was ActiView5, developed also by BioSemi. It provides a way of checking the electrode 113 impedance and the general quality of the incoming data. It was carefully checked that the impedance of each electrode was less 114 than 40 Ω before starting any recording session. Only a digital 208 Hz low-pass filter was used during acquisition time (no 115 high-pass filter was used). 116 Once the recording of each session was finished, a .bdf file was created and stored in computer PC2. This file contains the 117 continuous recording of the 128 EEG channels, the 8 external channels and the tagged events. 118 BCI Interaction Conditions 119 The design of the dataset was made having in mind as main objectives the decoding and understanding of the processes 120 involved in the generation of inner speech, as well as the analysis of its potential use in BCI applications. As described in the 121 “Background & Summary” Section, the generation of inner speech involves several complex neural networks interactions. With 122 the objective of localizing the main activation sources and analyzing their connections, we asked the participants to perform the 123 experiment under three different conditions: inner speech, pronounced speech and visualized condition. 124 Inner speech 125 Inner speech is the main condition in the dataset and it is aimed to detect the brain’s electrical activity related to a subject’s 126 thought about a particular word. In the inner speech runs, each participant was indicated to imagine his/her own voice, repeating 127 the corresponding word until the white circle turn blue. The subject was instructed to stay as still as possible and not to move 128 the mouth nor the tongue. For the sake of natural imagination, no rhythm cue was provided. 129 Pronounced speech 130 Although motor activity is mainly related to the imagined speech paradigm, inner speech may also show activity in the motor 131 regions. The pronounced speech condition was proposed with the purpose of finding motor regions involved in the pronunciation 132 matching those activated during the inner speech condition. In the pronounced speech runs, each participant was indicated 133 to repeatedly pronounce aloud the word corresponding to each visual cue. As in the inner speech runs, no rhythm cue was 134 provided.

2https://www.biosemi.com/products.htm 3https://www.biosemi.com/pics/cap_128_layout_medium.jpg 4https://es.parkerlabs.com/signagel.asp 5https://www.biosemi.com/software_biosemi_acquisition.htm 135 Visualized condition 136 Since the selected words have a high visual and spatial component, with the objective of finding any activity related to that being 137 produced during inner speech, the visualized condition was proposed. It is timely to mention that the main neural centers related 138 with this spatial thinking are located in the occipital and parietal regions41. In the visualized condition runs, the participant was 139 indicated to focus on mentally moving the circle shown in the center of the screen in the direction indicated by the visual cue. 140 Data Processing 141 In order to recast the continuous raw data into a more compact dataset and to facilitate their use, a transformation procedure 142 was proposed. Such processing was implemented in Python, mainly using the MNE library42, and the code along with the raw 143 data are available, so any interested reader can easily change the processing setup as desired (see Code Availability Section). 144 Raw data loading 145 A function that rapidly allows loading of the raw data corresponding to a particular subject and session, was developed. The raw 146 data stored in the .bdf file contains records of the complete EEG and external electrodes signals as well as the tagged events. 147 Events checking and correction 148 The first step of the signal processing procedure was checking for correct tagging of events in the signals. Missing tags 149 were detected and a correction method was proposed. The method detects and completes the sequences of events. After the 150 correction, no tags were missing and all the events matched those sent from PC1. 151 Re-reference 152 A re-reference step of the data to channels EXG1 and EXG2 was applied. This eliminates both noise and data drift, and it was 153 applied using the specific MNE re-reference function.

The data were filtered with a zero-phase bandpass finite impulse response filter using the corresponding MNE function. The lower and upper bounds were set to 0.5 and 100 Hz, respectively. This broad band filter aims to keep the data as raw as possible, allowing future users the possibility of filtering the data in their desired bands. A Notch filter in 50Hz was also applied. 158 Epoching and decimation 159 The data were decimated four times, obtaining a final sampling rate of 254 Hz. Then, the continuous recorded data were 160 epoched, keeping only the 4.5s length signals corresponding to the time window between the beginning of the concentration 161 interval and the end of the relaxation interval. The matrices of dimension [channels x samples] corresponding to each trial, 162 were stacked in a final tensor of size [Trials x channels x samples]. 163 Independent Components Analysis 164 Independent Components Analysis (ICA) is a standard and widely used blind source separation method for removing artifacts 165 from EEG signals43–45. For our dataset, ICA processing was performed only on the EEG channels, using the MNE implementa166 tion of the infomax ICA46. No Principal Component Analysis (PCA) was applied and 128 sources were captured. Correlation 167 with the EXG channels was used to determine the sources related to blink, gaze and mouth movement, which were neglected in 168 the process of reconstructing the EEG signals, for obtaining the final dataset. 169 EMG Control 170 The EMG control aims to determine whether a participant moved his/her mouth during the inner speech or visualized condition 171 runs. The simplest approach to find EMG activity is the single threshold method47. The baseline period was used as a basal 172 activity. The signals coming from the EXG7 and EXG8 channels were rectified and bandpass filtered between 1 and 20 Hz48–50. 173 The power in a sliding window of 0.5 s length with a time step of 0.05 s was calculated as implemented in Peterson et al51. The 174 power values were obtained by the following equation, where x[·] denotes the signal being considered, and s, S are the initial and final samples of the window, respectively. For every window, the computed powers were stacked and their mean and standard deviations were calculated and used to construct a decision threshold: th = mean(StackedPowerBaseline) + γ ∗ std(StackedPowerBaseline). ( 1 ) ( 2 ) 178 In Equation 2, γ is an appropriately chosen parameter. According to Micera et al.52 γ = 3 is a reasonable choice. The same 179 procedure was repeated for both channels and the mean power in the action interval of every trial was calculated. Then, if 180 one of those values, for either the EXG7 or EXG8 channels was above the threshold, the corresponding trial was tagged as 181 “contaminated”. 182 A total of 115 trials were tagged as contaminated, which represents a 2.5% of the inspected trials. The number of tagged 183 trials is shown in Table 1. The tagged trials and their mean power corresponding to EXG7 and EXG8 were also stored in 184 a report file. In order to reproduce the decision threshold, the mean and standard deviation power for the baseline for the 185 corresponding session were also stored in the same report file. 186 The developed script performing the control is publicly available and interested readers can use it to conduct different 187 analyses with the single threshold method.

After session 1, subject sub-03 claimed that, due a missinterpretation, he/she performed only one inner speech run and three visualized condition runs. The condition tag was appropriately corrected.

Data Records 192 All data files can be accessed at repository53. All files are contained in a main folder called “Inner Speech Dataset”, structured 193 as depicted in Figure 4, organized and named using the EEG data extension of BIDS recommendations54, 55. The final dataset 194 folder is composed of ten subfolders containing the session raw data, each one corresponding to a different subject. There is an 195 additional folder, containing five files obtained after the proposed processing: EEG data, Baseline data, External electrodes data, 196 Events data and a Report file. We now proceed to describe the contents of each one of these five files along with the raw data. 197 Raw data 198 The raw data file contains the continuous recording of the entire session for all 136 channels. The mean duration of the 199 recordings is 1554 seconds. The .bdf file contains all the EEG/EXG data and the tagged events with further information about 200 the recording sampling rate, the names of the channels and the recording filters, among other information. The raw events are 201 obtained from the raw data file and contain the tags sent by PC1, synchronized with the recorded signals. Each event code, its 202 ID and description are depicted in Table 2. A spurious event, of unknown origin, with ID 65536 appeared at the beginning of 203 the recording and also it randomly appeared within some sessions. This event has no correlation with any sent tag and it was 204 removed in the “Events Check” step of the processing. The raw events are stored in a three column matrix, where the first 205 column contains the time stamp information, the second has the trigger information, and the third column contains the event ID. 206 EEG data 207 Each EEG data file, stored in .fif format, contains the acquired data for each subject and session, after processing as described 208 above. Each one of these files contains an MNE Epoched object, with the EEG data information of all trials in the corresponding 209 session. The dimension of the corresponding tensor data is [Trials x 128 x 1154]. The number of trials changed among 210 participants in each session, as explained in the “Data Aquisition” Section. The number of channels used for recording was 128 211 while the number of samples was 1154, each one of them corresponding to 4.5 s of signal acquisition with a final sampling rate 212 of 256 Hz. A total of 1128 pronounced speech trials, 2236 inner speech trials and 2276 visualization condition trials, were 213 acquired, distributed as shown in Table 4. 214 External electrodes data 215 Each one of the EXG data files contains the data acquired by the external electrodes after the described processing was applied, 216 with the exception of the ICA processing. They were saved in .fif format. The corresponding data tensor has dimension [Trials 217 x 8 x 1154]. Here, the number of EXG trials equals the number of EEG data trials, 8 corresponds to the number of external 218 electrodes used, while 1154 corresponds to the number samples of 4.5 s of signal recording at a final sampling rate of 256 Hz. 219 Events Data 220 Each event data file (in .dat format) contains a four column matrix where each row corresponds to one trial. The first two 221 columns were obtained from the raw events, by deleting the trigger column (second column of the raw events) and renumbering 222 the classes 31, 32, 33, 34 as 0, 1, 2, 3, respectively. Finally, the last two columns correspond to condition and session number, 223 respectively. Thus, the resulting final structure of the events data file is as depicted in Table 5. 224 Baseline data 225 Each baseline data file (in .fif format) contains a data tensor of dimension [1 x 136 x 3841]. Here, 1 corresponds to the 226 only recorded baseline in each session, 136 corresponds to the total number of EEG + EXG channels (128+8), while 3841 227 corresponds to the numbers of seconds of signal recording (15) times the final sampling rate (256 Hz). Through a visual 228 inspection it was observed that the recorded baselines of subject sub-03 in session 3 and subject sub-08 in session 2, were 229 highly contaminated. 230 Report 231 The report file (in .pkl format) contains general information about the participant and the particular results of the session 232 processing. Its structure is depicted in Table 3. 233

Technical Validation 234 Attentional Monitoring 235 The evaluation of the participant’s attention was performed on the inner speech and the visualized condition runs. It was aimed 236 to monitor their concentration on the requested activity. The results of the evaluation showed that participants correctly followed 237 the task, as they performed very few mistakes (Table 6; mean ± std = 0.5 ± 0.62). Subjects sub-01 and sub-10 claimed that 238 they had accidentally pressed the keyboard while answering the first two questions in session 1. Also, after the first session, 239 subject sub-01 indicated that he/she felt that the questions were too many, reason for which, for the subsequent participants, the 240 number of questions was reduced, in order to prevent participants from getting tired. 241 Event Related Potentials 242 It is well known that Events Related Potentials (ERPs) are manifestations of typical brain activity produced in response to 243 certain stimuli. As different visual cues were presented during our stimulation protocol, we expected to find brain activity 244 modulated by those cues. Moreover, we expected this activity to have no correlation with the condition nor with the class 245 and to be found across all subjects. In order to show the existence of ERPs, an average over all subjects, for each one of the 246 channels at each instant of time, was computed using all the available trials (Nave = 5640), for each one of the 128 channels. 247 The complete time window average, with marks for each described event is shown in Figure 5. Between t = 0.1 s and t = 0.2 s 248 a positive-negative-positive wave appears, as it is clearly shown in Figure [5-A]. A similar behavior is observed between t = 0.6 249 s and t = 0.7 s, but now with a more pronounced potential deflection, reflecting the fact that the white triangle (visual cue) 250 appeared at t = 0.5 s (see Figure [5-B]). At time t = 1 s, the triangle disappeared and only the white fixation circular remained. 251 As shown in Figure [5-C], a pronounced negative potential followed. It is reasonable to believe that this negative potential is the 252 so-called “Contingent Negative Variation” ERP, which is typically related to the “warning-go” stimuli56. The signal appears 253 to be mostly stable for the rest of the action interval. As seen in Figure [5-D], a positive peak appears between t = 3.8 s and 254 t = 3.9 s, in response to the white circle turning blue, instant at which the relax interval begins. 255 Time-Frequency Representation 256 With the objective of finding and analyzing further differences and similarities between the three conditions, a Time-Frequency 257 Representation (TFR) was obtained by means of a wavelet transform, using the Morlet Wavelet. The implementation is available 258 in the file “TFR_representations.py”, at our GitHub repository (see Code Availability Section). 259 Inter Trial Coherence 260 By means of the TFR, the Inter Trial Coherence (ITC) was calculated for all 5640 trials (all together). A stronger coherence 261 was found within the concentration, cue and relax intervals, mainly at lower frequencies (see Figure 7). Also, the beginning of 262 the action interval presents a strong coherence. This could be a result of the modulated activity generated by the disappearance 263 of the cue. 264 Now, instead of taking the ITC of all trials (all together) we calculated the ITC for all the trials belonging to each one of 265 the three conditions, separately. Of the three conditions, pronounced speech appears to have a more intense global coherence, 266 mainly at lower frequencies. This is most likely due to the fact that there seems to exist a quite natural pace in the articulation 267 of generated sounds. Inner speech and visualized condition show consistently lower coherence during the action interval (see 268 Figures 7-A and 7-C). All these findings are consistent with the ERPs found in the time domain. 269 Averaged Power Spectral Density 270 Using all available trials for each condition, the Averaged Power Spectral Density (APSD) between 0.5 and 100 Hz was 271 computed. This APSD is defined as the average between all PSDs of the 128 channels. Figure 8 shows all APSD plots, in 272 which shaded areas correspond to ±1 std of all channels. As shown in the Inter Trial Coherence Section, all trials have a strong 273 coherence up to t = 1.5 s. Therefore, comparisons were made only in the action interval between 1.5 and 3.5 s. As it can 274 be seen, the plots in Figure 8 show a peak in the alpha band [8 - 12 Hz] for all conditions, as it was to be expected, with a 275 second peak in the beta band [12 - 30 Hz]. Also, pronounced speech shows higher power at high frequencies (beta-gamma), 276 which is most likely related to the brain motor activity and muscular artifacts. Finally, a narrow depression at 50 Hz appears, 277 corresponding to the Notch filter applied during data processing. 278 Spatial Distribution 279 In order to detect regions where neural activity between conditions is markedly different, the power difference in the main 280 frequency bands between each pair of conditions, was computed. As in the Averaged Power Spectral Density section, the time 281 window used was 1.5 - 3.5 s. The Power Spectral Density (PSD) was added to the analysis to further explore regions of interest. 282 Shaded areas on the PSD graphics in Figure 9 corresponds to ±1 std of the different channels used. No shaded area is shown 283 when only one channel was used to compute the PSD. 284 The top row of Figure 9 shows a comparison between inner and pronounced speech. In the alpha band, a major inner speech 285 activity can be clearly seen in the central occipital/parietal region. The PSD was calculated using channels A4, A5, A19, A20 286 and A326 and shows a difference of approximately 1 dB at 11 Hz. On the other hand, in the beta band, the spatial distribution 287 of the power differences shows an increased temporal activity for the pronounced condition, consistent with muscular activity 288 artifacts. Here, the PSD was calculated using channels B16, B22, B24 and B29 for the right PSD plot and channels D10, D19, 289 D21 and D26 for the left PSD plot. Pronounced speech shows higher power in the whole beta band with a more prominent 290 difference in the central left area. 291 The middle row of Figure 9 shows a comparison of the pronounced speech against the visualized condition. In the alpha 292 band, the visualized condition presents a larger difference in the central parietal regions and a more subtle difference in the 293 lateral occipital regions. The PSD was calculated using channels A17, A20, A21, A22 and A30. Here again, a difference 294 of about 1 dB at 11 Hz can be observed. In the beta band, an intense activity in the central laterals regions appears for the 295 pronounced condition. For this band, the PSD was calculated using the same channels as in the comparison between inner and 296 pronounced speech for the beta band. As seen, power for pronounced speech is higher than for the visualized condition in the 297 whole beta band, mainly in the left central region. This result is consistent with the fact that the occipital region is related to the 298 visual activity while the central lateral region is related to the motor activity. 299 Finally, a comparison of the inner speech with the visualized condition is shown in the bottom row of Figure 9. Visualized 300 condition exhibits a stronger activity in the laterals occipital regions in both the alpha and beta bands. This was to be expected 301 since the visualized condition, containing a stronger visual component, generates marked occipital activity. Interesting, inner 302 speech shows a broad although subtle higher power in the alpha band in a more parietal region. For the alpha band, the PSDs 303 were computed using channels A10 and B7 for the left and right plots respectively. In both plots, the peak corresponding to the 304 inner speech condition is markedly higher than the one corresponding to the visualized condition. For the beta band, the PSD 305 was calculated using channels A13 and A26 for the left and right PSD plots, respectively. As it can be observed, the power for 306 the visualized condition in the whole beta band is higher than the inner speech power. It is timely to mention that no significant 307 activity was presented in the central regions for neither of both conditions. 308

Usage Notes 309 The processing script was developed in Python 3.757, using the MNE-python package v0.21.042, NumPy v1.19.258, Scipy 310 v.1.5.259, Pandas v1.1.260 and Pickle v4.061. The main script, “InnerSpeech_processing.py”, contains all the described 311 processing steps and it can be modified to obtain different processing results, as desired. In order to facilitate data loading and 312 processing, six more scripts defining functions are also provided. 313 The stimulation protocols were developed using Psychtoolbox-339 in MatLab R2017b40. The auxiliary functions, including 314 the parallel port communication needed to send the tags from PC1 to BioSemi Active 2, were also developed in MatLab. The 315 execution of the main script, called “Stimulation_protocol.m”, shows the visual cue in the screen to the participant, and sends, 316 via parallel port, the event being shown. The parallel port communication was implemented in the function “send_value_pp.m”. 317 The main parameter that has to be controlled in the parallel communication is the delay needed after sending each value. This 318 delay allows the port to send and receive the sended value. Although we used a delay of 0.01 s, it can be changed as desired for 319 other implementations. 320

Code Availability 321 In line with reproducible research philosophy, all codes used in this paper are publicly available and can be accessed at 322 https://github.com/N-Nieto/Inner_Speech_Dataset. The stimulation protocol and the auxiliary MatLab 323 functions are also available. The code was run in PC1, and shows the stimulation protocol to the participants while sending the 324 event information to PC2, via parallel port. The processing Python scripts are also available. The repository contains all the 325 auxiliary functions to facilitate the load, use and processing of the data, as described above. By changing a few parameters in 326 the main processing script, a completely different process can be obtained, allowing any interested user to easily build his/her 327 own processing code. Additionally, all scripts for generating the TFR and the plots here presented, are also available. 6BioSemi nomenclature for a head cap with 128 channels - https://www.biosemi.com/pics/cap_128_layout_medium.jpg 329 330 331 375 28. Suppes, P., Han, B. & Lu, Z.-L. Brain-wave recognition of sentences. Proc. Natl. Acad. Sci. 95, 15861–15866 (1998). 376 29. Pasley, B. N. et al. Reconstructing speech from human auditory cortex. PLoS Biol. 10 (2012). 377 30. Cheung, C., Hamilton, L. S., Johnson, K. & Chang, E. F. The auditory representation of speech sounds in human motor 378 cortex. eLife 5, e12577 (2016). 379 31. Mitchell, T. M. et al. Predicting human brain activity associated with the meanings of nouns. Science 320, 1191–1195 380 (2008). 381 32. Huth, A. G., De Heer, W. A., Griffiths, T. L., Theunissen, F. E. & Gallant, J. L. Natural speech reveals the semantic maps 382 that tile human cerebral cortex. Nature 532, 453–458 (2016). 383 33. Pressel-Coreto, G., Gareis, I. E. & Rufiner, H. L. Open access database of EEG signals recorded during imagined speech. 384 In 12th International Symposium on Medical Information Processing and Analysis (SIPAIM) (2016). 385 34. Kaya, M., Binli, M. K., Ozbay, E., Yanar, H. & Mishchenko, Y. A large electroencephalographic motor imagery dataset for 386 electroencephalographic brain computer interfaces. Sci. Data 5, 180211 (2018). 387 35. Ofner, P. et al. Attempted arm and hand movements can be decoded from low-frequency EEG from persons with spinal 388 cord injury. Sci. Reports 9, 1–15 (2019). 389 36. Ofner, P., Schwarz, A., Pereira, J. & Müller-Putz, G. R. Upper limb movements can be decoded from the time-domain of 390 low-frequency EEG. PLoS ONE 12, e0182578 (2017). 391 37. Tangermann, M. et al. Review of the BCI competition IV. Front. Neurosci. 6, 55 (2012). 392 38. Höhne, J. et al. Motor imagery for severely motor-impaired patients: Evidence for brain-computer interfacing as superior 393 control solution. PLoS ONE 9, 1–11, 10.1371/journal.pone.0104854 (2014). 394 39. Brainard, D. H. The psychophysics toolbox. Spatial vision 10, 433–436 (1997). 395 40. MATLAB. version 7.10.0 (R2010a) (The MathWorks Inc., Natick, Massachusetts, 2010). 396 41. Kandel, E. R. et al. Principles of neural science, vol. 5 (McGraw-hill New York, 2000). 397 42. Gramfort, A. et al. MNE software for processing MEG and EEG data. Neuroimage 86, 446–460 (2014). 398 43. Jung, T.-P. et al. Extended ICA removes artifacts from electroencephalographic recordings. Adv. Neural Inf. Process. Syst. 399 894–900 (1998). 400 44. Vorobyov, S. & Cichocki, A. Blind noise reduction for multisensory signals using ICA and subspace filtering, with 401 application to EEG analysis. Biol. Cybern. 86, 293–303 (2002). 402 45. Makeig, S., Bell, A. J., Jung, T.-P. & Sejnowski, T. J. Independent component analysis of electroencephalographic data. In 403 Advances in Neural Information Processing Systems, 145–151 (1996). 404 46. Bell, A. J. & Sejnowski, T. J. An information-maximization approach to blind separation and blind deconvolution. Neural 405 Comput. 7, 1129–1159 (1995). 406 47. Thexton, A. A randomisation method for discriminating between signal and noise in recordings of rhythmic electromyo407 graphic activity. J. Neurosci. Methods 66, 93–98 (1996). 408 48. Porcaro, C., Medaglia, M. T. & Krott, A. Removing speech artifacts from electroencephalographic recordings during overt 409 picture naming. NeuroImage 105, 171–180 (2015). 410 49. Laganaro, M. & Perret, C. Comparing electrophysiological correlates of word production in immediate and delayed naming 411 through the analysis of word age of acquisition effects. Brain Topogr. 24, 19–29 (2011). 412 50. Ganushchak, L. Y. & Schiller, N. O. Motivation and semantic context affect brain error-monitoring activity: an event-related 413 brain potentials study. Neuroimage 39, 395–405 (2008). 414 51. Peterson, V., Galván, C., Hernández, H. & Spies, R. A feasibility study of a complete low-cost consumer-grade brain415 computer interface system. Heliyon 6, e03425 (2020). 416 52. Micera, S., Vannozzi, G., Sabatini, A. & Dario, P. Improving detection of muscle activation intervals. IEEE Eng. Medicine 417 Biol. Mag. 20, 38–46 (2001). 418 53. Nieto, N., Peterson, V., Rufiner, H., Kamienkowski, J. & Spies, R. "Inner Speech", http://doi.org/10.18112/openneuro. 419 ds003626.v1.0.1 (2021). 420 54. Gorgolewski, K. J. et al. The brain imaging data structure, a format for organizing and describing outputs of neuroimaging 421 experiments. Sci. Data 3, 1–9 (2016).