SACM: SEEG-Audio Contrastive Matching for Chinese Speech Decoding

Overview

This repository contains the official implementation of the SACM framework for Chinese speech decoding, as described in our paper "SACM: SEEG-Audio Contrastive Matching for Chinese Speech Decoding".

Dataset

The experiment uses a corpus of 48 Mandarin Chinese monosyllabic words. The word list can be found in A_Preprocessing/labels.py. The dataset is available upon request.

Installation

1. Clone the repository:

git clone https://github.com/WangHongbinary/SACM.git
cd SACM

2. Install dependencies:

pip install -r requirements.txt

Usage

Data Preprocessing

Process SEEG and audio data:

python A_Preprocessing/preprocess_SEEG.py

Configuration parameters can be adjusted in A_Preprocessing/preprocess_config.py.

Feature Extraction

Extract audio features:

python C_Decoding/dataset/wav_SEEG.py

Model Training & Evaluation

Speech Detection

Run the following script to train and evaluate the speech detection model:

cd B_Detect/run
bash run_exp.sh

Results will be saved as CSV files in the log directory.

Speech Decoding

Execute the following to perform speech decoding:

cd C_Decoding/run
bash run_exp.sh

Results will be saved as CSV files in the log directory.

Additional Analysis

The repository includes additional code for:

• Speech decoding on VocalMind: D_Decoding_VocalMind/
• Statistical analysis: E_Stats_test/
• Figure generation: F_Figure_code/

Project Structure

.
├── A_Preprocessing/      # Data preprocessing scripts
├── B_Detect/             # Speech detection implementation on HUST-MIND
├── C_Decoding/           # Speech decoding implementation on HUST-MIND
├── D_Decoding_VocalMind/ # Speech decoding implementation on VocalMind
├── E_Stats_test/         # Statistical analysis
└── F_Figure_code/        # Figure generation scripts

Citation

Cite our paper:

[Citation information will be added upon publication]