Ionio Podcast Rag Pipeline - Video Downloading, Transcription & RAG System

A complete video processing pipeline that downloads YouTube videos, transcribes them using OpenAI's Whisper, and provides a RAG (Retrieval-Augmented Generation) interface using OpenAI's Responses API for querying transcripts.

Features

• 📺 YouTube Video Download: Download videos from YouTube channels or individual URLs
• 🎵 Audio Extraction: Convert videos to high-quality audio files
• 📝 AI Transcription: Transcribe audio using OpenAI's Whisper with speaker diarization
• 🤖 RAG System: Query transcripts using OpenAI's Responses API with file search
• 🌐 Web Interface: Streamlit-based chat interface for interacting with transcripts
• 📊 Progress Tracking: Real-time progress bars for all operations

Setup

1. Create and activate virtual environment:

python3 -m venv venv
source venv/bin/activate  # On Unix/macOS
# or
.\venv\Scripts\activate  # On Windows

2. Install dependencies:

pip install -r requirements.txt

3. Install FFmpeg:

• macOS: brew install ffmpeg
• Ubuntu/Debian: sudo apt-get install ffmpeg
• Windows: Download from https://ffmpeg.org/download.html

4. Set up environment variables: Create a .env file in the project root:

# Hugging Face token for speaker diarization
HF_TOKEN=your_huggingface_token_here

# OpenAI API key for RAG system
OPENAI_API_KEY=your_openai_api_key_here

# Optional: Custom Whisper prompt
WHISPER_PROMPT="This is a conversation involving topics about AI, machine learning, and technology."

Usage

1. Download YouTube Videos

# Download from a channel
python download_youtube.py --channel CHANNEL_ID

# Download single video
python download_youtube.py --url "https://www.youtube.com/watch?v=VIDEO_ID"

# Download with custom output directory
python download_youtube.py --channel CHANNEL_ID --output-dir my_videos

2. Transcribe Videos

# Transcribe all videos in the videos directory
python transcribe.py

# Transcribe specific audio file
python transcribe.py --input path/to/audio.wav

# Simple transcription without speaker diarization
python transcribe.py --simple

3. Run RAG System

# Start the Streamlit web interface
streamlit run app.py

Then open your browser to the displayed URL (usually http://localhost:8501) to:

• Upload transcript files to OpenAI's file search
• Chat with your transcripts using natural language
• Get source-cited responses from your video content

Project Structure

TranscribeRohan/
├── setup.sh                 # Quick setup script  
├── app.py                   # Streamlit RAG interface
├── rag_system.py            # OpenAI Responses API integration
├── transcribe.py            # Main transcription script
├── extract_audio.py         # Audio extraction utilities
├── download_youtube.py      # YouTube downloader
├── requirements.txt         # All dependencies
├── .env                     # Environment variables (create this)
├── transcripts/             # Generated transcript files (JSON)*
├── audio/                   # Extracted audio files (WAV)*
├── videos/                  # Local video files*
└── downloaded_videos/       # YouTube downloads*

*Directories are created automatically when needed

Output Formats

Transcripts

Saved as JSON files in transcripts/ directory:

Simple transcription:

[
  {
    "start": "00:00:00",
    "end": "00:00:05", 
    "text": "Transcribed text here"
  }
]

With speaker diarization:

[
  {
    "start": "00:00:00",
    "end": "00:00:05",
    "speaker": "SPEAKER_1", 
    "text": "Transcribed text here"
  }
]

Dependencies

• Core: Python 3.8+, FFmpeg
• AI Models: OpenAI Whisper, Pyannote.audio for speaker diarization
• APIs: OpenAI API for RAG functionality
• Web: Streamlit for user interface
• Media: yt-dlp for YouTube downloads, ffmpeg-python for audio processing

Notes

• Automatic Directory Creation: All necessary directories (transcripts/, audio/, videos/, downloaded_videos/) are created automatically when running scripts or setup
• Large media files (videos/audio) are excluded from git by default
• Transcript JSON files in transcripts/ are preserved in git
• The RAG system uses OpenAI's latest Responses API with file search capabilities
• Speaker diarization requires a Hugging Face account and token

Troubleshooting

• SSL Certificate Issues: The app handles SSL certificate verification automatically
• Memory Issues: For large files, consider processing videos in smaller batches
• API Rate Limits: OpenAI API calls are automatically rate-limited
• File Upload Limits: OpenAI has file size limits for uploaded documents