VocalMind: A Stereotactic EEG Dataset for Vocalized, Mimed, and Imagined Speech in Tonal Language

🚀 Getting Started

Prerequisites

1. Download the dataset

   • Get it from Zenodo
   • Place the downloaded data under the /data/ directory

2. Download HiFi-GAN Vocoder model checkpoints:

   • Get it from Google Drive
   • Place it under vocoder/checkpoints

Data Processing

The repository includes scripts for processing original sEEG and audio data into processed data:

• Audio Processing: Check extract_audio_tokens function in src/audio_preproc.py
• EEG Processing: Check preprocess_eeg function in src/eeg_preproc.py

Training

You can use the processed data in the dataset to directly train the model. To train the model, run src/train_cross_val.py