Deepfake Speech Detection Project

This project focuses on detecting deepfake speech using a combination of Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN).

Project Outline

1. Data Preparation

   • Audio files are organized into a directory structure with 'data/fake' and 'data/real' subdirectories.
   • The generateDataCSV.py script is used to generate CSV files for organizing the audio dataset into training, validation, and evaluation sets.

2. Data Preprocessing

   • The train1.py script preprocesses the audio files to extract MFCC features.
   • MFCC features are saved to disk for future use.

3. Model Training

   • The train1.py script defines a CNN-RNN model and trains it on the preprocessed data.
   • The trained model is evaluated on the validation and test data.

4. Model Evaluation

   • The eval.py script evaluates the trained model on the test data.

5. Running the Application

   • The app.py script uses the trained model to classify audio files and creates a web-based user interface using Streamlit.

Step-by-Step Guide

1. Prepare Data

1. Place your audio dataset in the following directory structure:

   /path/to/root/dataset/
   ├── data
   │   ├── fake
   │   └── real
   ├── generateDataCSV.py
   ├── train1.py
   ├── eval.py
   └── app.py
   

2. Run the generateDataCSV.py script to generate CSV files for organizing the audio dataset:

   python generateDataCSV.py

   • The generated CSV files will be saved in the csvFilesReduced directory, as evaluate train and validate.csv.

2. Train the Model

1. Run the train.py script to train the model:

   python train.py

1.1. You may also Run the trainProcessedSample.py script to train the model, here features are already extracted of a large dataset:

python trainProcessedSample.py

3. Evaluate the Model

1. Run the eval.py script to evaluate the trained model:

   python eval.py

4. Run the Application

1. Run the app.py script to start the application:

   streamlit run app.py

Notes

• This is a basic model implementation. Feel free to modify and enhance it based on your requirements and dataset.

---