Imagined Speech in Spanish: EEG Dataset Acquisition Protocol and Baseline Classification Results

Abstract

This work presents a new Spanish-language electroencephalography (EEG) dataset for imagined speech, designed to support research in braincomputer interface (BCI) applications for assistive communication. A structured experimental protocol was developed to guide the acquisition process, incorporating auditory comprehension, imagined speech, and articulated speech production stages to enhance cognitive engagement and enable signal validation. The dataset includes 16 participants (9 male and 7 female), each performing 14 linguistic prompts consisting of nine words and five vowels. EEG signals were recorded using an open-source, low-cost acquisition system (OpenBCI Cyton + Daisy) with 16 channels configured according to the international 1020 system. The collected signals were preprocessed, segmented, and evaluated through a deep learning classification framework adapted from recent imagined speech decoding approaches. Five classification experiments were conducted to assess the discriminability of the imagined speech signals. The results showed accuracies above the chance level across all experiments, achieving 30.79% 4.76 for vowel classification, 20.81% 3.11 for word classification, and up to 74.61% 7.11 for binary wordvowel discrimination. Comparisons with public datasets demonstrated that the proposed dataset achieves competitive or superior performance despite using low-cost hardware. The code used in this work is available at https://github.com/GracielaRamirezA/Imagined-Speech-in-Spanish.git.