Generates podcasts from the web.
The Vision
What if you could turn any topic into an engaging podcast episode? Feed in a URL, get back a conversation between two AI voices discussing the content.
What I Built
An all-in-one tool that:
- Scrapes content from any URL
- Summarizes and structures key points
- Generates a two-voice dialogue script
- Produces audio using AI voices
- Outputs a ready-to-publish podcast episode
Stack: Python, NotebookLM, ElevenLabs, Whisper
Key Learnings
1. Two voices > one voice
Monologue podcasts are hard to listen to. Dialogue creates natural rhythm, lets you explore ideas from multiple angles, and keeps attention.
2. Scripts need editing
AI-generated scripts sound robotic without human polish. The 80/20 is: AI writes the draft, human adds personality and flow.
3. Audio quality is table stakes
Bad audio = instant skip. Even the best content fails with poor production. AI voices have gotten good enough that quality is achievable without a studio.
4. Curation beats generation
The best episodes come from carefully selected sources, not random URLs. Garbage in, garbage out.
Agent Quick Start
# Virtual Podcast Studio
Turn any topic into a podcast episode with AI voices.
## Pipeline
1. Input: URL or text
2. Summarize content
3. Generate dialogue script
4. Text-to-speech conversion
5. Output: Audio file
## Tools Used
- NotebookLM (dialogue generation)
- ElevenLabs or similar (TTS)
- Whisper (optional transcription)
## Stack
Python, Various AI APIs
## Links
- Repo: https://github.com/sergiopesch/virtualpodcaststudio