Table of Contents

1. Project title
2. Project introduction
   • 2.1. Overview
   • 2.2. Context and Objectives
3. Dataset description
   • 3.1. Experiment Configuration
   • 3.2. Recording Methodology
   • 3.3. Download the Dataset
   • 3.4. Contributing
4. Event Codes
   • 4.1. Each Class
   • 4.2. Start and End of the Experiment
   • 4.3. Channel Labels
5. Reference

1. Project title

Robust Decoding of Multi-class Imagined Classification

2. Project Introduction

2.1. Overview

EEG recordings captured the imagined vocalization of five distinct speech words/phrases. There's a total of 350 trials, with 70 trials for each class. Out of these, 60 trials per class are designated for training and 10 trials per class for validation. The testing data, comprising 10 trials for each class, will be disseminated in the future. The data has been segmented according to the cues provided (event markers).

2.2. Context and Objectives

The primary objective of this research is to achieve reliable classification of multiple imagined speech classes. The imagined vocalizations correspond to five essential communication words/phrases: ‘hello’, ‘help me’, ‘stop’, ‘thank you’, and ‘yes’. Throughout the study, participants were comfortably positioned in a chair, facing a 24-inch LCD display. They were guided to vividly imagine pronouncing the provided word as if they were actually verbalizing it, ensuring no physical movement or sound production. Participants were strictly advised to focus only on the given task and not engage in any other cognitive activities. Furthermore, they were cautioned to remain still and minimize eye blinks during both the imagination phase and cue reception. During all imagination trials, participants only saw a black screen, ensuring no external stimuli interfered with their brain activity.

3. Dataset description

3.1. Experiment Configuration

Auditory prompts of the five words/phrases played for 2 seconds, succeeded by a cross mark displayed for a duration between 0.8-1.2 seconds. As the cross mark vanished, participants commenced their imagined vocalization. This cycle of the cross mark display and imagined speech was repeated four times for each randomized cue. Following this, a 3-second relaxation period was allocated, allowing participants to reset before the next prompt.

3.2. Recording Methodology

EEG data collection was facilitated by a BrainAmp EEG signal amplifier, produced by BrainProduct GmbH, Germany. The data was captured using the BrainVision software and processed using MATLAB 2019a. Recording was done using 64 EEG electrodes based on the 10-20 international setup. The grounding and reference electrodes were situated at Fpz and FCz respectively, and the impedance between the sensors and the participants' scalp was consistently kept below 15 kΩ. The dataset comprises 300 training trials, 50 validation trials, and an upcoming test set of 50 trials for subjects 1–15.

3.3. Download the Dataset

To obtain the EEG Imagined Speech Dataset, click on the link below:

Download EEG Imagined Speech Dataset

Please cite our work or give appropriate credits if you use the dataset for your research or any other purposes.

3.4. Contributing

If you'd like to contribute to this dataset or have any suggestions, please open an issue or submit a pull request.

4. Event Codes

4.1. Each Class

Class	Event Code
Hello	1
Help me	2
Stop	3
Thank you	4
Yes	5

4.2. Start and End of the Experiment (a Whole Session)

Class	Event Code
Start	13
End	14

4.3. Channel Labels

Channel No.	Channel Label	Channel No.	Channel Label	Channel No.	Channel Label	Channel No.	Channel Label
1	Fp1	16	T8	31	O2	46	FT10
2	Fp2	17	TP9	32	PO10	47	C5
3	F7	18	CP5	33	AF7	48	C1
4	F3	19	CP1	34	AF3	49	C2
5	Fz	20	CP2	35	AF4	50	C6
6	F4	21	CP6	36	AF8	51	TP7
7	F8	22	TP10	37	F5	52	CP3
8	FC5	23	P7	38	F1	53	CPz
9	FC1	24	P3	39	F2	54	CP4
10	FC2	25	Pz	40	F6	55	TP8
11	FC6	26	P4	41	FT9	56	P5
12	T7	27	P8	42	FT7	57	P1
13	C3	28	PO9	43	FC3	58	P2
14	Cz	29	O1	44	FC4	59	P6
15	C4	30	Oz	45	FT8	60	PO7

5. Reference

J. H. Jeong et al., "2020 International brain–computer interface competition: A review," Frontiers in Human Neuroscience, Vol. 16, 2022, p. 898300.