🛡️ The Sonic Shield: Quantum-Entropic Audio Defense

A cryptographic audio watermarking system designed to poison Generative AI voice cloning models (e.g., ElevenLabs, HiFi-GAN) using True Quantum Randomness harvested from IBM's 127-qubit processors.

---

🚨 The Problem

Generative AI voice cloning poses a significant security risk for identity theft and disinformation. Most current audio watermarking defenses rely on Pseudo-Random Number Generators (PRNGs). Because PRNGs are deterministic, sophisticated attackers can reverse-engineer the "seed," predict the noise pattern, and use phase-cancellation to strip the watermark.

⚛️ The Solution: Quantum Entropy

The Sonic Shield abandons math.random() in favor of True Quantum Randomness (QRNG). By executing Hadamard gates on IBM Quantum hardware, we force qubits into superposition. Measuring these states yields a probabilistic collapse that is physically non-deterministic.

This entropy is injected into audio files using a Multi-Layered DSP Strategy that remains inaudible to humans but fatal to AI feature extraction.

🏛️ Architecture

The defense consists of three distinct layers:

1. Layer 1: Ultrasonic Shield (>18kHz)
   • High-amplitude noise injected above the Nyquist limit of human hearing.
   • Feature: Modulated with a Morse Code Ownership Signature (e.g., "Q-S-E-C") for cryptographic attribution.
2. Layer 2: Sub-Perceptual Grit (<4kHz)
   • Low-amplitude noise buried inside the vocal frequency range.
   • Purpose: Protects against "Downsampling Attacks" (converting audio to 16kHz to strip Layer 1).
3. Layer 3: Quantum Phase Poisoning
   • Uses the Quantum Key to stochastically rotate the phase angle of the audio signal by $\pi/4$.
   • Effect: Disrupts the phase alignment required by neural vocoders to reconstruct clear speech.

---

📸 Visual Proof

1. Spectrogram Analysis

Top: Original | Middle: Protected | Bottom: The Injected Quantum Noise

2. Quantum Phase Scattering

Left: Structured Clean Phase | Right: Quantum-Scattered Phase (Poisoned)

---

🛠️ Installation

Option A: Local Python

1. Clone the repository:

   git clone https://github.com/yourusername/sonic-shield.git
   cd sonic-shield

2. Install dependencies (Requires FFmpeg and libsndfile):

   pip install -r requirements.txt

3. Configure IBM Quantum Token: Copy .env.example to .env and add your API token.

---

🚀 Usage

1. Generate a Quantum Key

Harvest entropy from IBM's Brisbane Processor (run once, use forever).

python main.py --mode keygen

2. Protect an Audio File

Inject the defense layers into your recording.

# Local
python main.py -i inputs/my_voice.wav --strength 0.015

3. Verify Ownership (Decoder)

The system includes a bandpass analyzer that listens to the ultrasonic range to detect the specific Morse Code signature embedded in the shield.

python src/decode.py

---

🧪 Scientific Validation

The project includes a verify module to ensure the protection is robust yet invisible.

• Signal-to-Noise Ratio (SNR): Consistently achieves >45dB (Industry standard for transparency is 35dB).
• Shannon Entropy: Quantum Keys demonstrate an entropy of 0.999+ (vs ~0.6 for weak PRNGs).
• Red Team Test: Included scripts simulate ffmpeg downsampling and MP3 compression attacks to verify Layer 2 survivability.

🧰 Tech Stack

• Quantum Computing: Qiskit SDK, IBM Quantum Runtime (SamplerV2).
• DSP: Librosa, SciPy (Butterworth Filters, STFT/ISTFT).
• Visualization: Matplotlib, NumPy.

📄 License

Distributed under the MIT License. See LICENSE for more information.

---

Built by CyberAsim. A Portfolio Project bridging Quantum Mechanics and Cybersecurity.