DeWave: Discrete EEG Waves Encoding for Brain Dynamics to Text Translation

This repository implements the paper DeWave: Discrete EEG Waves Encoding for Brain Dynamics to Text Translation by Duan et al. (2024).

Project Structure

• conformer.py: Implementation of neural network model based on Conformer architecture
• VQVAE.py: Implementation of Vector Quantized Variational Autoencoder
• data.py: Data preprocessing and loading code
• config.py: Model and training configuration
• train_decoding.py: Code for training the decoding model
• model_decoding.py: Implementation of the decoding model

Key Features

• EEG signal processing and analysis
• Time series data processing using Conformer architecture
• Feature extraction and compression through VQ-VAE
• EEG signal decoding model training

Installation and Usage

1. Install required packages:

pip install -r requirements.txt

2. Train the model:

python3 train_decoding.py --model_name BrainTranslator \
    --task_name task1_task2_taskNRv2 \
    --two_step \
    --pretrained \
    --not_load_step1_checkpoint \
    --num_epoch_step1 35 \
    --num_epoch_step2 30 \
    --train_input EEG \
    -lr1 0.0005 \
    -lr2 0.000005 \
    -b 32 \
    -s ./checkpoints/decoding