🔇 AI Audio/Video Censor

Automated Profanity & Content Censorship Tool — Powered by Whisper ASR

An AI-powered command-line tool that automatically detects and censors profanity, abusive, violent, and confidential content from audio and video files. Built using OpenAI's Whisper for speech-to-text transcription with word-level timestamps for precise, frame-accurate censorship.

---

🏗️ Architecture

┌──────────────────────────────────────────────────────────┐
│                    INPUT (Video/Audio)                    │
└────────────────────────┬─────────────────────────────────┘
                         │
                         ▼
┌──────────────────────────────────────────────────────────┐
│              STAGE 1: Audio Extraction                    │
│  ┌─────────────┐                                         │
│  │ pydub/FFmpeg │ → Extract audio → Convert to WAV (16kHz)│
│  └─────────────┘                                         │
└────────────────────────┬─────────────────────────────────┘
                         │
                         ▼
┌──────────────────────────────────────────────────────────┐
│              STAGE 2: Speech Transcription                │
│  ┌─────────────┐                                         │
│  │   Whisper    │ → Word-level timestamps                 │
│  │  (ASR Model) │ → [{word, start, end}, ...]            │
│  └─────────────┘                                         │
└────────────────────────┬─────────────────────────────────┘
                         │
                         ▼
┌──────────────────────────────────────────────────────────┐
│            STAGE 3: Content Classification                │
│  ┌──────────────────┐  ┌───────────────────┐             │
│  │ Keyword Matching  │  │ Contextual Analysis│            │
│  │ (Profanity List)  │  │ (Regex Patterns)   │            │
│  └────────┬─────────┘  └────────┬──────────┘             │
│           └──────────┬──────────┘                         │
│                      ▼                                    │
│          Flagged Segments [{word, start, end, category}]  │
└────────────────────────┬─────────────────────────────────┘
                         │
                         ▼
┌──────────────────────────────────────────────────────────┐
│              STAGE 4: Censorship Engine                   │
│  ┌──────────┐  ┌──────────┐                              │
│  │  BLEEP   │  │   MUTE   │                              │
│  │(1kHz tone)│  │(Silence) │                              │
│  └──────────┘  └──────────┘                              │
│                      │                                    │
│            Merge censored audio with video (FFmpeg)       │
└────────────────────────┬─────────────────────────────────┘
                         │
                         ▼
┌──────────────────────────────────────────────────────────┐
│           OUTPUT (Censored Video/Audio + Report)          │
└──────────────────────────────────────────────────────────┘

---

✨ Features

• 🎤 Whisper-Powered ASR — Accurate transcription with word-level timestamps
• 🔍 Two-Tier Detection — Keyword matching + contextual pattern analysis
• 🔇 Bleep or Mute — Choose between tone overlay or silent censoring
• 📊 Detection Report — JSON report with all flagged content & timestamps
• 🎬 Video Support — Process video files (extracts audio, censors, reassembles)
• 🎚️ Sensitivity Levels — Low / Medium / High detection sensitivity
• 📝 Custom Word Lists — Add your own profanity/banned words
• 🖥️ Rich CLI — Beautiful terminal output with progress bars and tables
• 🔌 Modular Design — Each component is independent and extensible

---

🚀 Quick Start

Prerequisites

• Python 3.8+
• FFmpeg (required for video files and non-WAV audio formats)
  • Windows: choco install ffmpeg or download from ffmpeg.org
  • macOS: brew install ffmpeg
  • Linux: sudo apt install ffmpeg

Installation

# Clone the repository
git clone https://github.com/yourusername/ai-audio-video-censor.git
cd ai-audio-video-censor

# Create virtual environment (recommended)
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Usage

# Basic usage — bleep profanity in an audio file
python -m src.main -i input.wav

# Mute mode with high sensitivity
python -m src.main -i podcast.mp3 --mode mute --sensitivity high

# Process a video file with report generation
python -m src.main -i video.mp4 -o clean_video.mp4 --report

# Use a larger Whisper model for better accuracy
python -m src.main -i audio.wav --model small --mode bleep

# Use a custom word list
python -m src.main -i input.wav --wordlist my_words.txt

---

📋 CLI Options

Option	Default	Description
-i, --input	required	Input audio/video file path
-o, --output	auto	Output file path
--mode	bleep	Censorship mode: bleep or mute
--model	base	Whisper model: tiny, base, small, medium, large
--sensitivity	medium	Detection level: low, medium, high
--wordlist	built-in	Path to custom word list file
--report	off	Generate JSON detection report
--bleep-freq	1000	Bleep tone frequency (Hz)

Sensitivity Levels

Level	Keyword Match	Partial Match	Contextual Analysis
Low	✅	❌	❌
Medium	✅	✅	❌
High	✅	✅	✅

---

📁 Project Structure

ai-audio-video-censor/
├── README.md                    # This file
├── requirements.txt             # Python dependencies
├── setup.py                     # Package setup
├── LICENSE                      # MIT License
├── .gitignore                   # Git ignored files
├── config/
│   ├── config.py                # Centralized settings
│   └── profanity_words.txt      # Default word list (categorized)
├── src/
│   ├── __init__.py              # Package init
│   ├── main.py                  # CLI entry point
│   ├── audio_extractor.py       # Audio extraction from video/audio
│   ├── transcriber.py           # Whisper transcription engine
│   ├── content_classifier.py    # Content detection & classification
│   ├── censor_engine.py         # Bleep/mute audio processing
│   ├── pipeline.py              # Pipeline orchestrator
│   └── utils.py                 # Logging & helper utilities
├── output/                      # Default output directory
└── samples/                     # Sample files directory

---

🔧 Supported Formats

Audio

.wav, .mp3, .flac, .aac, .ogg, .wma, .m4a

Video

.mp4, .mkv, .avi, .mov, .webm, .flv, .wmv

---

📊 Sample Output

🔇 AI Audio/Video Censor
Automated Profanity & Content Censorship Tool
Powered by Whisper ASR | Built for OpenVINO GSoC

  Mode:        bleep
  Model:       base
  Sensitivity: medium
  Word List:   profanity_words.txt

──────────────── Pipeline Starting ────────────────
  Input:  D:\samples\interview.wav
  Output: D:\pipeline\output\interview_censored.wav

✓ Whisper model 'base' loaded successfully
✓ Transcription complete: 342 words detected, language: en
✓ Content classification complete: 7 flagged segments found

──────────────── Pipeline Complete ────────────────

      📊 Censorship Report
┌──────────────────┬─────────┐
│ Metric           │ Value   │
├──────────────────┼─────────┤
│ Total Words      │ 342     │
│ Flagged Segments │ 7       │
│ Censor Mode      │ BLEEP   │
│ Processing Time  │ 12.4s   │
│ Language         │ en      │
└──────────────────┴─────────┘

✅ Output saved: output/interview_censored.wav

---

🗺️ Future Roadmap (OpenVINO Integration)

This project serves as a proof-of-concept for the OpenVINO GSoC Project #39. Future enhancements planned:

1. OpenVINO Model Optimization — Convert Whisper to OpenVINO IR format for accelerated inference on Intel hardware
2. VLM Visual Understanding — Use Vision Language Models (via OpenVINO GenAI) to detect violent visual content in video frames
3. DL Streamer Integration — Build a GStreamer-based real-time processing pipeline using Intel DL Streamer
4. LLM-Based Classification — Replace keyword matching with an OpenVINO-optimized LLM for context-aware content understanding
5. Microservices Architecture — Deploy as containerized microservices for scalable production use
6. Fine-Tuning Support — Fine-tune content detection models on domain-specific data

---

🛠️ Tech Stack

Component	Technology
ASR Engine	OpenAI Whisper
Audio Processing	pydub, FFmpeg
Content Detection	Keyword matching, Regex patterns
CLI Interface	argparse, Rich
Language	Python 3.8+

---

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.

---

🙏 Acknowledgments

• OpenAI Whisper — Speech recognition model
• OpenVINO Toolkit — AI inference optimization
• pydub — Audio manipulation library
• Rich — Terminal formatting library

---

Built with ❤️