← Back to ML Projects
PhonexGuard – Deepfake Audio Detection Platform
AI Cybersecurity Solution
PyTorch
CNN (ResNet)
Librosa
FastAPI
PostgreSQL
Redis
Docker
Nginx
Project Overview
PhonexGuard is a cutting-edge deepfake audio detection platform designed to secure communications and authenticate audio integrity. It leverages advanced deep learning models to identify AI-generated or manipulated audio files with high precision.
Key Contributions & Architecture:
- Deep Learning Model: Engineered and trained a ResNet-based CNN model, using Librosa for feature extraction, achieving an impressive 99.8% accuracy across a robust dataset of 12,432+ real and AI-generated audio samples. Real-time inference operates seamlessly under 1.5 seconds.
- Scalable Backend Infrastructure: Developed a robust, production-grade FastAPI backend fortified with JWT/RBAC authentication to ensure secure access. Utilized PostgreSQL for persistent data storage and Redis for caching and session management.
- Containerization & Deployment: Fully containerized the application architecture using Docker and orchestrated reverse proxying and load balancing via Nginx for highly scalable deployment.
- Comprehensive Dashboard: Built an analytics dashboard featuring dynamic waveform and Mel-Spectrogram visualization, complete detection history logging, automated PDF reporting capabilities, and a dedicated admin/security monitoring panel.
