Large Spanish EEG

EEG: silent and perceive speech on 30 Spanish sentences
Large Spanish Speech EEG dataset

Authors
<ul>
<li>Carlos Valle</li>
<li>Carolina Mendez-Orellana</li>
<li>María Rodríguez-Fernández</li>
</ul>

Resources:
<ul>
<li>Code availaible at: https://github.com/cgvalle/Large_Spanish_EEG</li>
<li>Publication: Valle, C., Mendez-Orellana, C., Herff, C., & Rodriguez-Fernandez, M. (2024). Identification of perceived sentences using deep neural networks in EEG. Journal of neural engineering, 21(5), 056044. </li>
</ul>

Abstract:
Decoding speech from brain activity can enable communication for individuals with speech disorders. Deep neural networks have shown great potential for speech decoding applications, but the large data sets required for these models are usually not available for neural recordings of speech impaired subjects. Harnessing data from other participants would thus be ideal to create speech neuroprostheses without the need of patient-specific training data.
In this study, we recorded 60 sessions from 56 healthy participants using 64 EEG channels and developed a neural network capable of subject-independent classification of perceived sentences. We found that sentence identity can be decoded from subjects without prior training achieving higher accuracy than mixed-subject models.
The development of subject-independent models eliminates the need to collect data from a target subject, reducing time and data collection costs during deployment. These results open new avenues for creating speech neuroprostheses when subjects cannot provide training data.

Experiment Design:
We investigated the neural signals recorded using a 64-channel EEG system during speech perception and silent speech production tasks involving 30 daily use sentences in Spanish. The participants were instructed to listen to a spoken sentence from an audio recording and then silently repeat the sentence without any motor action.

The experimental design, a modified version of a previous study (Dash, et al), comprises four segments: rest, perception, preparation, and silent speech production. The rest segment lasted five seconds, presenting a fixation cross (+) before the stimulus onset.

During the perception section, the participants listened to the stimulus. Prior to subject S18, the perception section lasted 4 s, with each sentence being repeated 7 times. From subject S19 onward, the duration of the perception segment was increased to 5 s to match the duration of the silent speech production segment and the number of repetitions per sentence was decreased to 6 in order to maintain the overall length of the experiment. The preparation segment lasted one second and presented a blank screen, serving as a separation marker between the perception and silent speech production tasks. In the silent speech production segment lasting five seconds, subjects were instructed to silently repeat the previously heard stimulus without motor action only once. It is important to note that this study exclusively focuses on the outcomes derived from the speech perception task.

Trials for each block of the 30 stimuli were presented in a randomized order to prevent anticipation and learning biases.

Contact:
Please contact us at this e-mail address if you have any question: [e-mail]

脑电图（EEG）：基于30句西班牙语语句的静默言语生成与言语感知任务数据集
大型西班牙语言语脑电图数据集

作者：卡洛斯·巴列（Carlos Valle）、卡罗利娜·门德斯-奥雷利亚纳（Carolina Mendez-Orellana）、玛丽亚·罗德里格斯-费尔南德斯（María Rodríguez-Fernández）

资源：
1. 代码开源地址：https://github.com/cgvalle/Large_Spanish_EEG
2. 发表论文：Valle, C., Mendez-Orellana, C., Herff, C., & Rodriguez-Fernandez, M. (2024). 《基于深度神经网络的脑电图言语感知语句识别》，《神经工程期刊（Journal of Neural Engineering）》，21(5), 056044.

摘要：
从脑活动解码言语，可为言语障碍人群实现沟通提供可能。深度神经网络在言语解码任务中展现出巨大潜力，但这类模型训练所需的大规模数据集，通常难以通过言语障碍受试者的神经记录获取。因此，借助健康受试者的数据构建无需受试者专属训练数据的言语神经假体，将是理想方案。
本研究通过64通道脑电图（EEG）设备，招募56名健康受试者完成60次实验记录，并开发了可实现跨受试者分类的静默言语感知语句识别神经网络。实验结果表明，无需预先训练即可实现受试者的语句身份解码，且准确率优于混合受试者模型。
跨受试者模型的开发无需针对目标受试者采集数据，降低了部署阶段的数据收集时长与成本。本研究结果为无法提供训练数据的受试者构建言语神经假体开辟了新路径。

实验设计：
本研究针对30句日常西班牙语语句，在言语感知与静默言语生成任务中，采集64通道脑电图系统记录的神经信号。实验要求受试者先聆听音频录制的语句，随后静默复述该语句且不产生任何运动动作。
本实验设计改编自先前研究（Dash等人），包含四个阶段：静息期、感知期、准备期与静默言语生成期。静息期时长5秒，呈现固定十字（+）直至刺激呈现时刻。
感知期阶段，受试者聆听刺激语句。在S18受试者之前，感知期时长为4秒，每句语句重复7次。自S19受试者起，感知期时长增至5秒，以匹配静默言语生成期的时长，同时将每句语句的重复次数降至6次，以维持整体实验时长。准备期时长1秒，呈现黑屏，作为感知期与静默言语生成期的分隔标记。静默言语生成期时长5秒，受试者需仅静默复述一次此前聆听的刺激语句，且不得产生运动动作。需特别说明的是，本研究仅聚焦于言语感知任务的结果。
30个刺激语句的每组实验试次均以随机顺序呈现，以避免预期偏差与学习偏差。

联系方式：
如有任何疑问，请发送邮件至：[e-mail]