FAKeless Security - Technical Documentation & System Reference

1. Executive Summary

FAKeless Security is a high-fidelity mobile application engineered to combat the rising threat of audio deepfakes and unauthorized voice cloning. The system provides a multi-layered defense mechanism, combining cloud-based forensic analysis with localized, on-device machine learning (TFLite) to protect and verify digital audio assets.

---

2. System Architecture

graph TD
    User((User)) --> UI[Flutter UI Layer]
    
    subgraph Screens
        UI --> Home[Home Dashboard]
        UI --> Protect[Protect Screen]
        UI --> Detect[Attacker Simulator]
        UI --> History[Security History]
    end

    subgraph ServiceLayer
        Protect & Detect --> API[ApiService]
        Protect --> TFLite[TFLiteService]
        UI --> AI[AiAgentService - Gemini]
        History --> Storage[HistoryService]
        API --> Crypto[CryptoService]
    end

    subgraph External
        API --> CloudProtect[Protection API]
        API --> CloudClone[Cloning API]
        AI --> GoogleGemini[Google Gemini API]
    end

    subgraph LocalAssets
        TFLite --> Model[delta_universal.tflite]
        Storage --> Prefs[Shared Preferences]
    end

Loading

---

3. Screen-by-Screen Use Case Analysis

3.1 SplashScreen (splash_screen.dart)

• Use Case: Initial brand immersion and system health check.
• Logic: Verifies the presence of the .env file and initializes core services (TfliteService, CryptoService) before navigating to the onboarding or home screen.

3.2 OnboardingScreen (onboarding_screen.dart)

• Use Case: User education for first-time security analysts.
• Interaction: Progressive disclosure of features (Protection vs. Detection) using high-quality Lottie animations.

3.3 MainShell (main_shell.dart)

• Use Case: The "Master Control" of the application.
• Role: Manages the IndexedStack for seamless navigation and hosts the Floating AI Assistant, ensuring security help is always one tap away regardless of the current screen.

3.4 HomeScreen (home_screen.dart)

• Use Case: Security Status Hub.
• Features:
  • Hero Section: Quick-access button to start audio protection.
  • Live Stats: Real-time count of protected files from HistoryService.
  • Metric Education: Deep dive into PESQ (Quality) and STOI (Intelligibility) scores to help users understand security tradeoffs.

3.5 ProtectScreen (protect_screen.dart)

• Use Case: Active Audio Defensive Layer.
• Detailed Workflow:
  1. Input: User records audio (PCM 16k WAV) or uploads a file.
  2. Analysis: The TfliteService analyzes the audio and recommends a "Protection Scale" (1-5).
  3. Defense: User applies a watermark via the ApiService (Cloud) or TfliteService (Local Fallback).
  4. Verification: Generates a Protection Certificate and displays a side-by-side Waveform Comparison.

3.6 DetectScreen (detect_screen.dart)

• Use Case: Adversarial Attack Simulation (The "Red Team" tool).
• Detailed Workflow:
  1. Cloning Interface: Allows a user to input "Target Text."
  2. API Execution: Sends the source audio and text to a remote cloning model.
  3. Validation: Users listen to the "Cloned" output to verify if the protection applied in the previous step successfully disrupted the cloning process.

---

4. UI Component (Widget) Catalog

Component	File	Use Case / Logic
Floating Assistant	floating_ai_assistant.dart	Persistent Gemini chat UI. Handles voice commands and file uploads conversationally.
Wav Visualizer	waveform_visualizer.dart	Custom-painter based waveform. Used to audit the "Adversarial Noise" applied during protection.
A/B Comparison	ab_comparison_player.dart	Synchronized audio playback for comparing Original vs. Protected/Cloned audio.
Level Meter	level_meter.dart	Real-time decibel tracking to ensure recording quality is sufficient for ML analysis.
Protection Cert	protection_certificate.dart	Visual "Proof of Security" card displaying model version, scale, and time-thwarted stats.

---

5. Service Layer Specification

5.1 AiAgentService (The Brain)

• Technology: Google Gemini Pro API.
• Use Case: Interprets user intent. If a user says "Protect this audio at maximum level," the service identifies the "intent" and returns a structured action that the UI uses to trigger _submitProtection(5).

5.2 TfliteService (Edge Intelligence)

• Model: delta_universal.tflite (8-bit quantized).
• Use Case: Performs local signal processing. If the device is offline, this service applies a "Universal Watermark" layer directly to the WAV bytes without cloud dependency.

5.3 ApiService (Cloud Integration)

• Network: HTTPS Multipart requests.
• Endpoints:
  • /protect: Applies advanced adversarial perturbation.
  • /detect: Forensic analysis of audio segments.

5.4 HistoryService (Audit Trail)

• Persistence: SharedPreferences + JSON Serialization.
• Use Case: Maintains a permanent record of every protection session, allowing users to share or re-verify protected audio from weeks ago.

---

6. Technical Requirements & Deployment

6.1 Environment (.env)

The app requires an environment file bundled in assets:

• GEMINI_API_KEY: Required for the Chat Intelligence.
• PROTECTION_API_ENDPOINT: URL for the cloud protection model.

6.2 Permission Handlers

The system implements strict permission checks for:

• Permission.microphone: For recording audio.
• Permission.storage: For saving/loading .wav files.

---

7. Security & Data Privacy

• Encryption: CryptoService uses AES-256 to obfuscate local path identifiers in the preference store.
• Data Lifecycle: Voice recordings are treated as ephemeral. Any temporary file generated during the "Trim" or "Protect" process is stored in the TemporaryDirectory and cleaned up by the OS when space is needed.

---

8. Maintenance Guide

• Updating Models: Replace assets/models/delta_universal.tflite and update the version string in TfliteService.
• UI Refresh: Centralized styles are located in lib/core/theme.dart. Modernizing the app-wide look only requires updating the AppTheme class.

FAKeless Security v1.0.0
Documentation updated: April 2026