Voice Reconstruction from Brain Signals
The algorithm aimed at reconstructing voice from EEG during imagined speech.
Key Contributions- A generative model capable of extracting frequency characteristics and sequential information from neural signals to generate speech.
- Addressed the constraint of imagined speech-based BTS system lacking ground truth voice by employing a domain adaptation method.
- Demonstrated the potential of robust speech generation by training only several words or phrases, with the model showing capability to learn phoneme level information from brain signals.
- This work is currently accepted for presentation at AAAI 2023.
- This work is based on Neurotalk. We will continue to develop and extend from this foundational work..
EEG Imagined Speech Decoding Using Diffusion-based Learning
Decoding EEG signals for imagined speech has been a complex task, primarily due to the high-dimensional nature of the data and a low signal-to-noise ratio.
Key Contributions- Our study introduces Diff-E, a novel method that utilizes denoising diffusion probabilistic models (DDPMs) and a conditional autoencoder to address these challenges.
- We've found that Diff-E substantially outperforms traditional machine learning techniques and baseline models in terms of decoding accuracy.
- These findings indicate the potential effectiveness of DDPMs for EEG signal decoding, suggesting possible applications in the development of brain-computer interfaces that enable communication through imagined speech.
- This work is currently accepted for presentation at Interspeech 2023.
- This work is based on Diff-E. We will continue to develop and extend from this foundational work.
Comprehensive collection of EEG signal preprocess/analysis codes
- It primarily focuses on the analysis of EEG data.
- This folder features different versions for Motor Imagery (MI), Steady-State Visual Evoked Potential (SSVEP), and Event-Related Potential (ERP).
- It serves as an invaluable resource for neuroscience researchers and data scientists.
- This content is particularly useful for those interested in EEG data processing and brain-computer interfaces.
- This work is based on GigaScience. We will continue to develop and extend from this foundational work.
The open software package, designed for developing Brain-Computer Interfaces (BCIs) with various advanced pattern recognition algorithms.
- Example codes for Motor Imagination (MI), Event-Related Potential (ERP), and Steady-State Visually Evoked Potential (SSVEP) in the 'Examples' folder.
- The package also features 'BMI_modules' with implementation functions, 'GUI module' for Graphic User Interface functions, and 'Paradigm' functions using Psychtoolbox.
- Additionally, it contains codes from other BCI groups and an OpenBMI demo. For questions or more information, visit http://openbmi.org.
- This work is based on OpenBMI. We will continue to develop and extend from this foundational work.
Motor imagination brain signal analysis codes
- This folder provides comprehensive analysis codes for motor imagination, including pre-processing, feature extraction, classification, and evaluation modules.
- It includes codes for a basic motor imagination paradigm and example codes for setting up experiments and conducting analysis.
- For inquiries, refer to the website http://openbmi.org.
- This work is based on TNNLS. We will continue to develop and extend from this foundational work.
This folder contains code for the topic 'Reconstructing Sentences from Brain Signals using Contextual and Semantic Information'. The corresponding paper is currently submitted for review to SMC 2023. It will be updated in the near future.
This OnlineDemo folder will continue to be updated for an online demo system that is currently under development.
