Goswami Whisper

A Flask web application for transcribing and searching Vedic lectures (Krishna consciousness lectures) with AI-powered tools.

Features

• Audio/Video Upload: Upload files or provide URLs (YouTube, etc.)
• Whisper Transcription: Accurate speech-to-text using OpenAI Whisper CLI
• AI Formatting: Intelligent formatting with GPT-4, optimized for Vedic/Sanskrit terminology
• MySQL Storage: Persistent storage of raw and formatted transcriptions
• Semantic Search: LaBSE-powered vector search using ChromaDB for RAG capabilities

Prerequisites

• Python 3.10+
• Docker and Docker Compose (for MySQL)
• Whisper CLI installed (pip install openai-whisper)
• NVIDIA GPU (recommended, the app is optimized for RTX 3090)
• FFmpeg (for audio processing)

Installation

1. Clone the repository:

   cd goswami-whisper
   

2. Start MySQL with Docker Compose:

   docker-compose up -d
   

3. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

4. Install dependencies:

   pip install -r requirements.txt
   

5. Configure environment variables:

   cp .env.example .env
   # Edit .env and set your OPENAI_API_KEY
   

6. Run the application:

   python main.py
   

7. Access the web interface: Open http://localhost:5000 in your browser

Configuration

Edit the .env file to configure:

• OPENAI_API_KEY: Your OpenAI API key for text formatting
• WHISPER_MODEL: Whisper model size (tiny, base, small, medium, large)
• MySQL credentials (default values work with Docker Compose)

Project Structure

goswami-whisper/
├── app/
│   ├── __init__.py          # Flask app factory
│   ├── models/
│   │   └── transcription.py  # SQLAlchemy models
│   ├── routes/
│   │   ├── main.py           # Home and view routes
│   │   ├── upload.py         # File/URL upload handling
│   │   └── search.py         # Vector search
│   ├── services/
│   │   ├── whisper_service.py    # Whisper CLI integration
│   │   ├── openai_service.py     # OpenAI API formatting
│   │   └── embedding_service.py  # LaBSE embeddings & ChromaDB
│   └── templates/            # Jinja2 HTML templates
├── uploads/                  # Uploaded files
├── chroma_db/               # ChromaDB vector storage
├── config.py                # Configuration
├── main.py                  # Application entry point
├── docker-compose.yml       # MySQL container
├── requirements.txt         # Python dependencies
└── .env                     # Environment variables

Usage

1. Upload a recording:

   • Go to the Upload page
   • Either upload an audio/video file or paste a YouTube URL
   • Wait for processing (transcription → formatting → indexing)

2. View transcriptions:

   • Browse recent transcriptions on the home page
   • Click to view full formatted and raw text

3. Search lectures:

   • Use the search bar in the navigation
   • Semantic search finds relevant content even with different wording

Technical Details

• Whisper: Uses OpenAI Whisper CLI with configurable model size
• OpenAI GPT-4: Formats transcriptions preserving Vedic/Sanskrit terminology
• LaBSE: Language-agnostic BERT embeddings for multilingual search
• ChromaDB: Local vector database for persistent embedding storage
• MySQL 8.0: Relational database for transcription metadata

License

MIT License