Voxie AI — Deepfake Authentication & Forensic Suite

Voxie is an advanced audio authenticity tool designed to detect AI-generated speech, verify spectral watermarks, and perform biological forensic analysis on voice recordings.

Built for high-stakes environments where audio integrity is paramount, Voxie provides a multi-layered security approach to expose even the most sophisticated "Synthetic Perfection" in modern deepfakes.

🚀 Live Demo: voxieai.streamlit.app

---

🛡️ Features

1. Multi-Layer AI Detection

Powered by the Wav2Vec2-L-v2 Deepfake Detector, Voxie analyzes the neural patterns of audio to identify synthetic signatures. It provides a real-time confidence score and classification (Real vs. Fake) for any uploaded WAV or MP3 file.

2. Biological Forensic Analysis (Layer 3)

Beyond simple AI models, Voxie extracts raw biological signal markers:

• Pitch Jitter: Detects the "Unnatural Stability" found in AI voices.
• Spectral Flux: Measures the rate of change in the power spectrum to identify synthetic synthesis.
• Harmonics-to-Noise Ratio (HNR): Detects "Synthetic Perfection" where the voice is too clean to be human.

3. Spectral Watermarking

A secure FFT-based watermarking system that allows you to "Sign" your audio.

• Embed: Inject a hidden, high-frequency spectral signature into any audio file.
• Verify: Instantly detect if a file carries a valid Voxie signature, ensuring it originated from a trusted source.

4. Neo-Brutalist Dashboard

An immersive, interactive "Bento-style" dashboard designed with a raw, high-contrast aesthetic:

• Dotted Grid Background: A technical, blueprint-style workspace.
• Interactive Interactivity: Cards lift and shift shadows on hover for a tactile experience.
• Typography: Powered by Bricolage Grotesque for a modern, aggressive tech feel.

5. Historical Analytics

All analyses are faithfully persisted to a local SQLite database, allowing for long-term trend visualization, risk distribution charts, and forensic history tracking.

---

🏗️ Architecture

• Frontend (Streamlit): Custom-built UI using CSS-in-Python for Neo-Brutalist styling. Implements responsive "Bento" layouts and interactive glassmorphism components.
• Backend (Python/PyTorch): Utilizes transformers for AI inference and librosa / scipy for deep signal processing and forensic feature extraction.
• Signal Pipeline: Raw binary audio is resampled to 16kHz, processed through an FFT spectral analyzer, and passed to the Wav2Vec2 transformer network.
• Database (SQLite): A lightweight, persistent storage engine for tracking every forensic operation performed.

---

⚙️ Setup & Installation

Prerequisites

• Python: v3.9 or higher.
• FFmpeg: Required for audio resampling (Install via brew install ffmpeg or apt install ffmpeg).

1. Clone & Environment

git clone https://github.com/ramanverse/Voxie.git
cd Voxie
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

2. Install Dependencies

pip install -r requirements.txt

3. Run Application

streamlit run app.py

---

🚀 Usage Guide

1. Dashboard: View your system health, total threats detected, and average risk scores.
2. Verify Audio:
   • Upload an audio file or use the provided samples.
   • Forensic Breakdown: Expand the metrics to see raw Jitter and HNR values.
   • Verdict: View the final risk level (Low, Medium, or High).
3. Embed Watermark: Upload a raw file to inject a secure spectral signature for future authentication.
4. History & Analytics: Track your past detections and visualize risk trends over time.

---

🛠️ Troubleshooting

• First Run Delay?: The app downloads a ~1.2GB AI model from Hugging Face on the first run. Please ensure a stable internet connection.
• Audio Error?: Ensure ffmpeg is installed on your system path.
• Database Locked?: Ensure only one instance of the app is running if you are performing heavy database operations.

---

Voxie AI — Authenticating the world's voice, one signal at a time.