Dys-Locate: A System for Dysarthric Speech Detection and Transcription

Dys-Locate is a machine learning-based system designed to detect and transcribe dysarthric speech, enhancing accessibility for individuals with speech impairments. Leveraging the TORGO dataset and advanced speech processing techniques, Dys-Locate aims to provide accurate and reliable transcription solutions for better communication and understanding.

Features

• Automatic Dysarthria Detection: Identifies whether speech is dysarthric or not.
• Speech Transcription: Transcribes dysarthric speech with improved accuracy.
• Interactive Interface: A Streamlit-based web application for real-time detection and transcription.

Project Structure

• README.md - Project Documentation
• proj.ipynb - Jupyter notebook for the model building
• torgo - Dataset direcotry
• app.py - Streamlit interface for the application
• requirements.txt - Python dependencies
• dysarthia_detection_model.h5 - the final trained model
• mfcc_data.pkl - pickle file containing the MFCCs for future use
• processed_data.pkl - pickle file containing the preprocessed audio file

Getting Started

Prerequisistes

• Python 3.8 or above
• GPU for trianing (optional)
• TORGO dataset (download from TORGO website)

Note: You will have to download the dataset on your own.

Installation

1. Clone the repository:

git clone https://github.com/your-repo/Dys_Locate.git
cd Dys_Locate

2. Install the required dependencies:

pip install -r requirements.txt

Dataset preparation

1. Place the TORGO dataset in the same direcotry.
2. Run the Jupyter notebook after replacing the path of the TORGO dataset.

Usage

Training the Model

Train the dysarthia detection model using the Jupyter notebook.

Running the Streamlit app

Launch the Streamlit interface for real-time interaction:

streamlit run app.py

Results

• Detection accuracy: 95%

Future work:

• Expanding the dataset to include diverse dysarthria cases.
• Improving transcription for severe dysarthric cases.
• Integrating Dys-Locate into assistive devices and mobile applications.

Contributing

We welcome contributions! Please follow the contribution guidelines to submit issues or pull requests.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• The TORGO dataset creators for their invaluable resource.
• The open-source community for the tools and libraries used.