AudioReplayAttackDetection

Audio Replay Attack Detection is a Python project for detecting replay attacks in audio recordings using a Support Vector Machine (SVM) model. The repository contains tools for dataset management, model training, and prediction.

Dataset

The dataset directory contains audio recordings used for training and testing the SVM model. It includes 100 real and 100 replayed audio samples, each recorded in mono at a sampling rate of 20500 Hz. Additionally, the raw subdirectory provides raw real and replayed audio files that can be used to expand the dataset.

Utilities

The utilities directory houses Python scripts for managing and analyzing the audio dataset. These scripts provide functionalities such as checking audio dataset information and extracting audio file details.

Usage

Installation

To install the required packages in venv, run the following command:

pip install numpy pandas librosa scikit-learn joblib

Training and Testing

To train and test the SVM model, run main.py. This script utilizes the provided dataset to train the model and saves the trained model files in .pkl format.

python main.py

Prediction For making predictions on new audio recordings, utilize predict.py. This script loads the trained SVM model from the .pkl file and prompts the user to input the path of the audio file to be classified as real or replayed.

python predict.py

UI

The repository also includes a user interface (UI) [only the UI - still under work] for easy integration. To use the UI, run flaskApp.py and navigate to http://localhost:5000 in your web browser.

python flaskAPP.py

License

This project is licensed under the Eclipse Public License 2.0 (EPL-2.0) - see the LICENSE file for details.

Contributor

Feel free to contribute, report issues, or suggest improvements!