What's this all about❓

Thanks for asking. Well, this project arose from a personal need to stay up to date with what's happening in my country 🇿🇲. I'm generally terrible at keeping up with current affairs, so I figured I'd put my programming skills to good use and take advantage of the recent AI developments to build a tool that would help me stay informed. This is the result. I hope you find it useful too.

Listen
Zed News Podcast

How does this work❔

Every Monday to Friday in the afternoon, a cron job runs "in the cloud" to fetch the latest news from a number of sources (which are mentioned in each podcast episode).

Each news item is then summarized using a language model, and a Python script compiles everything into some kind of transcript for the podcast episode.

The transcript is then converted into an audio file using text-to-speech technology.

Next, some basic sound mixing is done, which includes addition of an opening and closing jingle. Podcast metadata is then added to the audio file. At this point, the episode is ready to be published.

All of the above processes are handled automatically, without any human intervention. I think the animation below pretty sums it up. It's kinda like a Rube Goldberg machine 😄!

And now, some fun facts❗

  • Only news items published on the same day are included in each episode.
  • The same news item can be on multiple news platforms. Because each news item is summarized independently, the same news item can be repeated in the same episode, though using different words. This is still a work in progress. Hopefully, I'll figure out a way to detect duplicates and consolidate them into a single news item.
  • No opinions are expressed in the podcast. The news items are presented as they are, without any commentary.
  • Summaries generated using LLMs may lack full context, nuance, and may inadvertently introduce biases or inaccuracies due to the limitations of the model and the potential for misinterpretation of input. That's why the links to the articles are there for you to go and check for yourself if you need to!
  • The text to speech technology is not perfect. Sometimes, the pronunciation of certain words (especially Zambian words, names, places, etc.) is not correct. After some experimenting, I found that, of all the English voices available, the one with better pronunciation is Ayanda from South Africa. I think there's work that needs to be done towards having a Zambian English locale, just as we have South African English, Australian English, etc. Perhaps you and I can do something about this? 🤔
  • Because the news items are presented as summaries, the podcast episodes are usually short, between 8 and 20 minutes long, depending on the number of news items on that day.
  • Links to the original news articles are included in the show notes, so you can read the full articles if you want to know more.
  • This whole thing is an experiment in automation and artificial intelligence. The best way to learn a craft is by building things, and that's what I'm doing here. I'm learning as I go, and I'm having a lot of fun doing it. I hope you enjoy listening to the podcast as much as I enjoy tinkering with the technology behind it.
  • The idea for this project was actually inspired by Hackercast, an AI-generated podcast summary of Hacker News.

Need a helping hand?

Perhaps you need someone with technical expertise to help you solve a problem?
Or maybe you want us to collaborate on some cool stuff?

Let's talk