Decoding speech perception from non-invasive brain recordings

Nature Machine Intelligence, Journal Year: 2023, Volume and Issue: 5(10), P. 1097 - 1107

Published: Oct. 5, 2023

Language: Английский

---

The speech neuroprosthesis

Nature reviews. Neuroscience, Journal Year: 2024, Volume and Issue: 25(7), P. 473 - 492

Published: May 14, 2024

Language: Английский

---

Subject-independent meta-learning framework towards optimal training of EEG-based classifiers

Neural Networks, Journal Year: 2024, Volume and Issue: 172, P. 106108 - 106108

Published: Jan. 6, 2024

Language: Английский

---

On the role of generative artificial intelligence in the development of brain-computer interfaces

BMC Biomedical Engineering, Journal Year: 2024, Volume and Issue: 6(1)

Published: May 2, 2024

Abstract Since their inception more than 50 years ago, Brain-Computer Interfaces (BCIs) have held promise to compensate for functions lost by people with disabilities through allowing direct communication between the brain and external devices. While research throughout past decades has demonstrated feasibility of BCI act as a successful assistive technology, widespread use outside lab is still beyond reach. This can be attributed number challenges that need addressed practical including limited data availability, temporal spatial resolutions signals recorded non-invasively inter-subject variability. In addition, very long time, development been mainly confined specific simple patterns, while developing other applications relying on complex patterns proven infeasible. Generative Artificial Intelligence (GAI) recently emerged an artificial intelligence domain in which trained models used generate new properties resembling available data. Given enhancements observed domains possess similar development, GAI employed multitude synthetic activity; thereby, augmenting activity. Here, brief review recent adoption techniques overcome aforementioned provided demonstrating achieved using EEG data, enhancing spatiotemporal resolution cross-subject performance systems implementing end-to-end applications. could represent means would transformed into prevalent thereby improving quality life disabilities, helping adopting emerging human-computer interaction technology general use.

Language: Английский

---

Open Vocabulary Electroencephalography-to-Text Decoding and Zero-Shot Sentiment Classification

Proceedings of the AAAI Conference on Artificial Intelligence, Journal Year: 2022, Volume and Issue: 36(5), P. 5350 - 5358

Published: June 28, 2022

State-of-the-art brain-to-text systems have achieved great success in decoding language directly from brain signals using neural networks. However, current approaches are limited to small closed vocabularies which far enough for natural communication. In addition, most of the high-performing require data invasive devices (e.g., ECoG). this paper, we extend problem open vocabulary Electroencephalography(EEG)-To-Text Sequence-To-Sequence and zero-shot sentence sentiment classification on reading tasks. We hypothesis that human functions as a special text encoder propose novel framework leveraging pre-trained models BART). Our model achieves 40.1% BLEU-1 score EEG-To-Text 55.6% F1 EEG-based ternary classification, significantly outperforms supervised baselines. Furthermore, show our proposed can handle various subjects sources, showing potential high-performance system once sufficient is available. The code made publicly available research purpose at https://github.com/MikeWangWZHL/EEG-To-Text.

Language: Английский

---

Automated Recognition of Imagined Commands From EEG Signals Using Multivariate Fast and Adaptive Empirical Mode Decomposition Based Method

IEEE Sensors Letters, Journal Year: 2022, Volume and Issue: 6(2), P. 1 - 4

Published: Jan. 13, 2022

In this letter, a novel automated approach for recognizing imagined commands using multichannel electroencephalogram (MEEG) signals is presented. The multivariate fast and adaptive empirical mode decomposition method decomposes the MEEG into various modes. slope domain entropy $L_1$ -norm features are obtained from modes of signals. machine learning models such as k -nearest neighbor, sparse representation classifier, dictionary (DL) techniques used command classification tasks. efficacy proposed evaluated public database input has achieved average accuracy values 60.72, 59.73, 58.78% DL model selected left versus right, up down, forward backward based categorization

Language: Английский

---

Adaptive LDA Classifier Enhances Real-Time Control of an EEG Brain–Computer Interface for Decoding Imagined Syllables

Brain Sciences, Journal Year: 2024, Volume and Issue: 14(3), P. 196 - 196

Published: Feb. 21, 2024

Brain-Computer Interfaces (BCIs) aim to establish a pathway between the brain and an external device without involvement of motor system, relying exclusively on neural signals. Such systems have potential provide means communication for patients who lost ability speak due neurological disorder. Traditional methodologies decoding imagined speech directly from signals often deploy static classifiers, that is, decoders are computed once at beginning experiment remain unchanged throughout BCI use. However, this approach might be inadequate effectively handle non-stationary nature electroencephalography (EEG) learning accompanies use, as parameters expected change, all more in real-time setting. To address limitation, we developed adaptive classifier updates its based incoming data real time. We first identified optimal (the update coefficient, UC) used Linear Discriminant Analysis (LDA) classifier, using previously recorded EEG dataset, acquired while healthy participants controlled binary syllable decoding. subsequently tested effectiveness optimization control Twenty performed two sessions imagery syllables, LDA randomized order. As hypothesized, led better performances than one task. Furthermore, were closely aligned both datasets, same These findings highlight reliability classifiers improvement can shorten training time favor development multi-class BCIs, representing clear interest non-invasive notably characterized by low accuracies.

Language: Английский

---

Speech imagery decoding using EEG signals and deep learning: A survey

IEEE Transactions on Cognitive and Developmental Systems, Journal Year: 2024, Volume and Issue: 17(1), P. 22 - 39

Published: July 19, 2024

Language: Английский

---

Learning to operate an imagined speech Brain-Computer Interface involves the spatial and frequency tuning of neural activity

Communications Biology, Journal Year: 2025, Volume and Issue: 8(1)

Published: Feb. 20, 2025

Brain-Computer Interfaces (BCI) will revolutionize the way people with severe impairment of speech production can communicate. While current efforts focus on training classifiers vast amounts neurophysiological signals to decode imagined speech, much less attention has been given users' ability adapt their neural activity improve BCI-control. To address whether BCI-control improves and characterize underlying dynamics, we trained 15 healthy participants operate a binary BCI system based electroencephalography (EEG) through syllable imagery for five consecutive days. Despite considerable interindividual variability in performance learning, significant improvement was globally observed. Using control experiment, show that continuous feedback about decoded is necessary learning occur. Performance associated broad EEG power increase frontal theta focal enhancement temporal low-gamma activity, showing an imagined-speech involves dynamic changes features at different spectral scales. These findings demonstrate combining machine human successful strategy enhance controllability.

Language: Английский

---

On the Role of Activation Functions in EEG-to-Text Decoder

Lecture notes in computer science, Journal Year: 2025, Volume and Issue: unknown, P. 46 - 60

Published: Jan. 1, 2025

Language: Английский

---