Cantonese Radio Show Generator

A system for generating dual-host Cantonese radio shows from PDF/text content, with planned real-time Q&A interruption capabilities.

Project Structure

• backend/ - FastAPI backend (Python)
• userstory.md - User stories and requirements
• system_architecture.md - System design and flow
• spec.md - Technical specifications
• Action_plan.md - Development phases
• prompt_design.md - LLM prompt engineering

Current Phase

Phase 1: The Radio - Basic broadcast functionality

• ✅ PDF/TXT upload and text extraction
• ✅ Script generation with dual-host dialogue (Agent A)
• ✅ Audio generation using Azure TTS
• ⏳ Real-time Q&A interruptions (Phase 2)
• ⏳ Polish and optimizations (Phase 3)

Quick Start

See backend/README.md for detailed setup instructions.

Windows Development (Recommended)

1. Install Docker Desktop for Windows
2. Clone this repository
3. Navigate to backend/ directory
4. Copy env_template.txt to .env and configure API keys
5. Run docker-compose up --build

Local Python Development

1. Install Python 3.11 or 3.12
2. Navigate to backend/ directory
3. Create virtual environment: python -m venv venv
4. Activate: venv\Scripts\activate (Windows) or source venv/bin/activate (Mac/Linux)
5. Install dependencies: pip install -r requirements.txt
6. Copy env_template.txt to .env and configure API keys
7. Run: uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

API Documentation

Once running, visit:

• API Docs: http://localhost:8000/docs
• Health Check: http://localhost:8000/health

Technology Stack

• Backend: FastAPI (Python)
• LLM: Azure OpenAI (GPT-4o) / OpenAI / Anthropic Claude
• TTS: Azure Cognitive Services Speech (Cantonese voices)
• Containerization: Docker

Development Notes

• Windows: Native x86_64 support - optimal for Azure Speech SDK
• Mac M1/M2/M3: May require Docker with x86_64 emulation or use OpenAI TTS as workaround
• Phase 1: Single-pass generation, no interruptions/Q&A yet

License

[Your License Here]