Created by two nerds ;)

This repository is not final and still in development. You're free to report any bugs and we'll try to roll-out an update as soon as possible. This project is to be presented at Regeneron ISEF 2025, to be held at Columbus, Ohio, USA.

VoiceAuth - AI Voice Detection System

A desktop application that uses machine learning to differentiate between AI-generated and human voices based on audio samples. VoiceAuth employs logistic regression and advanced audio feature extraction to provide accurate classification of voice samples.

Features

• ✅ Audio Processing: Extract features from various audio formats (WAV, MP3, OGG, FLAC)
• 🧠 Machine Learning: Logistic regression model with hyperparameter optimization
• 🔍 Real-time Classification: Analyze and classify voice samples instantly
• ⚡ Performance: Hardware acceleration through parallel processing
• 📊 Visualization: Feature importance and model performance metrics
• 🔄 Feedback System: Continuously improve model accuracy with user feedback
• 💾 Cross-Platform Support: Works on Windows, macOS, and Linux

Screenshots

Installation

Prerequisites

• Python 3.8 or higher
• C++ Build Tools (for Windows users)

Windows Setup

Step 1: Install C++ Build Tools (required for NumPy, SciPy, and other scientific packages)

1. Download Microsoft C++ Build Tools:

   • Go to Microsoft C++ Build Tools
   • Click "Download Build Tools"

2. Run the installer:

   • When prompted for workloads, select "C++ build tools"
   • In the right panel, ensure the following components are selected:
     • MSVC v142 - VS 2019 C++ x64/x86 build tools
     • Windows 10 SDK
     • C++ CMake tools for Windows
   • Click Install

Alternative: Use Anaconda/Miniconda

If you prefer to avoid installing build tools, you can use Anaconda/Miniconda which provides pre-compiled binaries:

1. Download and install Miniconda
2. Create an environment: conda create -n voiceauth python=3.8
3. Activate it: conda activate voiceauth
4. Install packages: conda install numpy pandas matplotlib scikit-learn
5. Then install the remaining packages with pip: pip install -r requirements.txt

Standard Setup

1. Clone the repository:

   git clone https://github.com/zohaiblazuli/VoiceAuth.git
   cd VoiceAuth
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Verify your setup:

   python check_setup.py
   

   This script will check if you have all necessary components installed and provide guidance on fixing any issues.

4. Run the application:

   python run_voiceauth.py
   

Usage

Running the Application

VoiceAuth features a user-friendly interface with several tabs:

1. Import Sample Tab: Import audio files for classification
2. Record Sample Tab: Record your voice directly for classification
3. Feedback Tab: Provide feedback on classification results to improve the model
4. Information Tab: View model statistics and performance metrics

Sample Dataset Structure

For training the model, organize your dataset as follows:

dataset_folder/
├── ai_generated/  (folder containing AI-generated voice samples)
│   ├── sample1.wav
│   ├── sample2.wav
│   └── ...
└── human/  (folder containing human voice samples)
    ├── sample1.wav
    ├── sample2.wav
    └── ...

Technical Details

Path Management System

VoiceAuth uses a robust path management system to ensure compatibility across different environments, particularly when shared via GitHub. The system:

• Automatically determines the base directory: Works whether running from source, as a compiled executable, or from any relative directory
• Standardizes path access: All file paths are accessed through utility functions, not hardcoded strings
• Creates necessary directories: Output and model directories are automatically created if they don't exist
• Cross-platform compatibility: Paths are normalized for the operating system in use

Key path utility functions:

• get_base_dir(): Gets the base application directory
• get_resource_path(relative_path): Gets the absolute path to any resource
• get_media_path(filename): Gets the path to media files
• get_output_path(filename): Gets the path to output files or directories
• get_model_path(filename): Gets the path to model files

Feature Extraction

VoiceAuth extracts various audio features using the librosa library:

• MFCCs (Mel-Frequency Cepstral Coefficients)
• Spectral Centroid, Contrast, Rolloff
• Zero Crossing Rate
• Chroma Features
• Spectral Bandwidth
• Tempo and Beat Features
• Mel Spectrogram

Machine Learning Model

• Preprocessing: Standard scaling for feature normalization
• Feature Selection: Optional using SelectFromModel
• Model: Logistic regression with hyperparameter tuning
• Evaluation: Uses accuracy, precision, recall, and F1-score

Feedback System

The application includes an adaptive learning system that:

• Collects user feedback on classification results
• Stores correctly labeled samples
• Automatically retrains the model when sufficient feedback data is collected
• Updates the model in real-time

Development

Project Structure

• voiceauth.py: Main application module
• ui_components.py: UI components and widgets
• simple_model.py: Machine learning model implementation
• audio_processor.py: Audio processing and feature extraction
• batch_process.py: Batch processing of audio samples
• tabs.py: Implementation of application tabs
• utils.py: Utility functions, including path management
• media/: Contains graphics and media assets
• output/: Contains generated files (features, models)

Extending the Application

To add new features:

1. For new UI components, add them to ui_components.py
2. For new tabs, extend the functionality in tabs.py
3. For new model features, modify simple_model.py
4. For additional audio processing, update audio_processor.py

Troubleshooting

Common Issues

• Media files not found: If you see errors about missing media files, ensure the media directory is in the same location as the application.
• Model loading errors: Make sure the model has been trained and the appropriate model files exist in the output directory.
• Audio recording issues: Check your microphone permissions and settings.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Licensing & Copyright

Please do mention us when showcasing this repository.

Acknowledgements

• librosa for audio feature extraction
• scikit-learn for machine learning components
• PyQt5 for the GUI framework