I built a new tool since the last update, fixed a couple of bugs in the tools I already had, and now have a complete production workflow running on my own software from start to finish. This episode covers all three.
Podcast Idiot Playout
The newest addition to the toolset is Podcast Idiot Playout, a radio-style music playback app. I used to run Play It Live on Windows for this kind of thing, but there wasn’t an equivalent on Linux after I switched over. Playout fills that gap by letting me queue up a sequence of tracks that fade into each other, the same way a radio station handles music breaks between segments.
Listen to this episode here
Get the podcast idiot podcast player
Right now I use it for two things. It plays my intro music for the Moderate Method podcast, and it lets me build short custom mixes for my grandchildren, which I export as MP3s and load onto their YOTO players. I’ve only tested Playout on my Linux machine so far and it’s still rough around the edges, but it does the job I built it for.
Bug fixes to the RSS Feed Builder and Transcriber
The RSS Feed Builder had a bug that only showed up when creating a brand new feed from scratch. Editing an existing feed worked fine, but a new feed came out with a couple of tags formed incorrectly. The feed loaded fine in my own podcast player, Podcast Addict, and Podverse, but Apple rejected it, which is what led me to the problem. That fix is live on the site now, and new feeds validate correctly.
The Transcriber had a similar story on Windows. It didn’t work the first time I tested it there, so I went back through it and got it running. I still need to test it on a Mac, so if anyone wants to try that and report back, I’d take the help. On my own Linux machine the Transcriber doesn’t pick up the GPU, since the hardware is about 12 years old, but it still runs fine on CPU. A 40 to 45 minute episode takes about 15 to 20 minutes to transcribe.
The full production workflow
With all the pieces working, I now have an entire episode pipeline built from my own tools plus a couple of AI chat sessions, with no dependence on any single paid service. I record in Audacity, trim the front and back, and normalize the volume to catch any peaking from getting too close to the mic. That gives me the finished MP3.
The MP3 goes into the Podcast Idiot Transcriber to produce a transcript. The transcript goes into the Podcast Idiot Chapter Creator. I don’t have an API key set up yet, so I use the Chapter Creator to generate a prompt, then paste that prompt along with the transcript into ChatGPT rather than Claude, since ChatGPT produces better cartoon images for my chapter art. ChatGPT returns a chapters JSON file along with one combined image containing a cartoon for every chapter.
From there I use GIMP to cut that combined image apart into individual chapter images and save them as JPEGs, since PNG files end up too large for this. Once I have the MP3, transcript, chapters file, and chapter images, I upload everything to the right folders on my web server. Then I open the RSS Feed Builder, which pulls the existing feed through a proxy running on my server, add the new episode, and upload the updated feed. That’s the whole process, and it’s the same one I’ll use going forward for this show.
Try the tools and send feedback
I’m the only person who has used most of these tools so far, so there are almost certainly rough edges I can’t see. Some of the touch targets in the podcast player app are still too small on phones, for example, and I’ve only fixed part of that. If you try any of the tools and run into something that doesn’t work or doesn’t make sense, send me an email at mail@hapamedia.com and let me know.
All of the tools are free to download, and since they were built through plain conversation with AI, you can take the code and have your own AI assistant modify it however you want. That’s part of the point of building things this way.
Leave a Reply