Why Wikipodia?
Back in 2005, more than 18 years ago, I had the idea to create a podcast from Wikipedia articles using text-to-speech (TTS). And – just in case – I reserved the domain wikipodia.de.
Yet, the TTS engines weren’t that great and licensing the generated audio was not affordable at all. On top, the texts of Wikipedia is nothing that you should pass to a TTS engine right away.
Fast forward to end of 2023, I cleaned up some of my domains and stumbled across wikipodia.de again. Technology evolved a lot and so I eventually took the Christmas break 23/24 to let AI bring Wikipodia to live.
From creating a podcast script from a wikipedia page over generating the voice to creating the cover art – all is done via OpenAI APIs. At a cost of a few cents.
Is it perfect? No. During a car trip with the family the ratings were between „boooooring“ and a more polite „not yet suitable for mainstream“.
It is a nice experiment. And as every new release can take advantage of the evolution of the underlying AI, I wonder how the quality changes over time. And what looking back to todays experiments will be like in 2042.
How the Podcast is made.
Wikipodia building blocks
Wikipodia was offline for one year. I sat down in the last week of May 2026 to let Codex review and update the Wikipodia toolchain. I also updated to the latest OpenAI models.
Until 05/2025
- manual article of the day list
- gpt-4o-mini for script generation
- DALL-E for cover Art
- gpt4o-mini-tts for audio generation
- sox to mix the sound files
- WordPress with Seriously Simple Podcasting
- hacky nodejs scripts
Starting 06/2026
- new article of the days scraping via Wikipedia API
- script generation updated to gpt-5-mini
- using gpt-image-2 for cover art generation
- use of new, better voice „marin“
- new logo (by ChatGPT)
FAQs
Who did this?
Stefan Motz
Father & Geek
