🛡️ Spazor: Real-time Multilingual Profanity Monitoring System

Spazor is a high-performance, distributed AI system designed to monitor environmental audio for profanity and toxic speech in real-time. By combining edge computing with a centralized AI processing server, Spazor transcribes multilingual speech, identifies specific speakers via voice biometrics, and triggers immediate alerts for violations.

---

🚀 Key Improvements & Features

• Ultra-Low Latency STT: Switched to Groq Whisper-large-v3 with in-memory audio processing. Buffer reduced to 0.75s, providing near-instantaneous transcription.
• Hindi & Hinglish Curse Detection: Integrated a custom LLM prompt for Groq (Llama-3) that is specialized in detecting regional Hindi slang, abuse, and code-switched profanity.
• Dynamic Speaker Management: A new sidebar utility allows administrators to map voice fingerprints (Speaker IDs) to real names, persisting them to a local SQLite database.
• Active Discord Integration: Automated notifications are dispatched to Discord via webhooks immediately upon violation detection, including the transcript and severity score.
• Start/Stop Controls: The dashboard now supports toggling the monitoring state to manage system resources effectively.
• Live DB Logging: All violations are dynamically stored and displayed in a real-time monitoring table.

---

🛠️ Architecture & Tech Stack

1. Capture & Streaming

• Hardware: Compatible with ESP32 (WROOM/S3) or local microphone inputs.
• Protocol: 16kHz Mono PCM streaming via WebSockets.
• Buffer Optimization: Uses 24,000-byte chunks (~0.75s) to balance context window and speed.

2. Intelligence Layer (AI Server)

• Transcription: Groq Cloud API (Whisper-large-v3) for speed, with local faster-whisper fallback.
• Analysis: Groq Llama-3-8b for high-accuracy multilingual toxicity detection.
• Biometrics: pyannote.audio for speaker identification and voice mapping.

3. Infrastructure

• Backend: FastAPI (Asynchronous WebSocket handling).
• Database: SQLAlchemy + SQLite for persistent user naming and violation history.
• Frontend: Streamlit with custom CSS (Glassmorphism, dynamic alert states).

---

📂 Project Structure

├── app/
│   ├── api/            # WebSocket logic & Audio processing
│   ├── services/       # AI Engines (Groq, Toxicity, Speaker ID, Alerts)
│   ├── models.py       # User & Log Database Schemas
│   ├── database.py     # SQLAlchemy Setup
│   └── config.py       # Environment Configuration
├── app_frontend.py     # Streamlit Dashboard (Glassmorphic UI)
├── requirements.txt    # Python dependencies
└── .env                # API Keys & Discord Webhooks

---

⚙️ Setup & Deployment

1. Installation

# Clone and enter repo
git clone https://github.com/SanjayGanesh614/Profanity-Detection-System.git
cd Profanity-Detection-System

# Setup Environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Configuration

Update your .env file:

GROQ_API_KEY=gsk_xxx...
DISCORD_WEBHOOK_URL=https://discord.com/api/webhooks/...
DATABASE_URL=sqlite:///./spazor.db

3. Execution

# Terminal 1: Start Backend
python -m app.main

# Terminal 2: Start Frontend
streamlit run app_frontend.py

---

🛡️ Security & Privacy

• Privacy First: Optional local inference mode using faster-whisper and local RoBERTa models for sensitive environments.
• Audit Trails: Every violation is logged with a timestamp, speaker fingerprint, and raw transcript for forensic review.

---

Maintained by: Sanjay Ganesh & Team