🎙️ Podsite - AI Podcast Generation

Transform web content and text into engaging AI-generated podcasts with natural-sounding conversations between two speakers.

🚀 Two Versions Available

📱 Regular Version (app.py)

• Uses OpenAI GPT-4 for script generation
• Requires OpenAI API key
• Fast and reliable cloud-based AI
• Pay-per-use pricing model

🔓 Open Source Version (app_oss.py)

• Uses Ollama with local AI models
• No API keys required for script generation
• Completely private and offline
• Free to use after setup

✨ Features

• 🌐 Website Scraping: Extract content from any URL using Firecrawl
• 📋 Text Input: Paste any text content directly
• 🎭 Multiple Styles: Choose from Conversational, Interview, Debate, or Educational formats
• ⏱️ Flexible Duration: Generate podcasts from 5 to 20 minutes
• 🎵 AI Audio Generation: Convert scripts to natural-sounding speech using Kokoro TTS
• 📥 Export Options: Download both scripts (JSON) and audio files (WAV)
• 📚 Source Management: Store multiple sources and generate podcasts from any of them

🚀 Quick Start

Prerequisites

• Python 3.11 or 3.12
• uv package manager

For Regular Version (app.py):

• OpenAI API key (required)
• Firecrawl API key (optional, for web scraping)

For Open Source Version (app_oss.py):

• Ollama installed and running
• A local AI model (e.g., gpt-oss:20b)

Installation

1. Navigate to the project:

cd ai-podcast-generation

2. Install dependencies using uv:

uv sync

3. Set up your environment variables:

cp .env.example .env

4. Add your API keys to .env (for regular version):

OPENAI_API_KEY=your_openai_api_key_here
FIRECRAWL_API_KEY=your_firecrawl_api_key_here  # Optional

Running the Apps

Regular Version (OpenAI):

uv run streamlit run app.py

Open Source Version (Ollama):

# First, ensure Ollama is running with your model
ollama serve
ollama pull gpt-oss:20b  # or your preferred model

# Then run the OSS app
uv run streamlit run app_oss.py

Both apps will open in your default browser at http://localhost:8501

📖 Usage

Step 1: Add Sources

Navigate to the "📁 Add Sources" tab:

Option A: Scrape Website

1. Select the "🌐 Website" tab
2. Enter a URL (e.g., https://example.com/article)
3. Click "Add Website"
4. Wait for the content to be scraped
5. Requires FIRECRAWL_API_KEY in .env

Option B: Paste Text

1. Select the "📋 Text" tab
2. Enter a source name
3. Paste your content
4. Click "Add Text"

Your sources will appear in the sidebar with:

• Source name/title
• Type (Website or Text)
• Word count
• Delete button (🗑️) to remove

Step 2: Generate Podcast

1. Navigate to the "🎙️ Studio" tab
2. Select a source from the dropdown
3. Choose your podcast style:
   • Conversational: Natural, friendly discussion
   • Interview: Q&A format
   • Debate: Different perspectives
   • Educational: Explanatory with clarifying questions
4. Choose duration (5, 10, 15, or 20 minutes)
5. Click "🎙️ Generate Podcast"
6. Wait for script and audio generation
7. Download the results!

🏗️ Project Structure

ai-podcast-generation/
├── app.py                          # Main Streamlit application
├── src/
│   ├── podcast/
│   │   ├── script_generator.py    # Podcast script generation
│   │   └── text_to_speech.py      # Audio generation with TTS
│   └── web_scraping/
│       └── web_scraper.py          # Web content extraction
├── outputs/                        # Generated audio files
├── pyproject.toml                  # Project dependencies (uv)
├── .env.example                    # Environment variables template
└── README.md                       # This file

🛠️ Technology Stack

• Streamlit: Web interface
• OpenAI GPT-4: Script generation via CrewAI
• Kokoro TTS: Natural text-to-speech synthesis
• Firecrawl: Web content extraction
• uv: Fast Python package management

📝 Example Output

The app generates:

1. Podcast Script (JSON format):

   • Structured dialogue between Speaker 1 and Speaker 2
   • Metadata including source, duration, and line count

2. Audio Files (WAV format):

   • Individual segments for each speaker turn
   • Complete combined podcast with natural pauses
   • High-quality 24kHz audio

🔧 Configuration

Podcast Settings

• Style: Conversational, Interview, Debate, Educational
• Duration: 5, 10, 15, or 20 minutes
• Configured in the Studio tab

Audio Settings

Modify in src/podcast/text_to_speech.py:

• Speaker voices (default: af_heart and am_liam)
• Sample rate (default: 24000 Hz)
• Pause duration between segments (default: 0.2s)

🐛 Troubleshooting

Web Scraping Not Available

If you see "Web scraping will not be available":

• Add FIRECRAWL_API_KEY to your .env file
• Get a key from https://www.firecrawl.dev
• Restart the app

TTS Not Available

If you see "TTS not available":

uv pip install kokoro>=0.9.4

OpenAI API Errors

• Check your API key is correctly set in .env
• Ensure you have sufficient API credits
• Verify your API key has access to GPT-4 models

Import Errors

Make sure all dependencies are installed:

uv sync

💡 Workflow

1. Add Source (URL or Text)
   ↓
2. Content is stored in session
   ↓
3. Select source in Studio
   ↓
4. Generate script (GPT-4)
   ↓
5. Generate audio (Kokoro TTS)
   ↓
6. Download script & audio

📄 License

This project is part of the AI Engineering Hub.

🙏 Acknowledgments

• Inspired by NotebookLM's podcast generation feature
• Uses Kokoro TTS for natural-sounding speech synthesis
• Powered by OpenAI's language models
• Web scraping by Firecrawl

🤝 Contributing

---

Built with ❤️ using Streamlit and AI