Adventures in AI: personal podcast generator

Music and audio is a big passion of mine; I worked as a Sound Engineer for several years before becoming a Software Engineer, and played in bands for many years too.

OpenAI’s recent announcement on their audio tooling got me itching to play with what’s available.

For a while I’ve wanted to learn about things via audio whilst on the move — it’s a good way to get an overview of something without ploughing through an article or a textbook.

Orra

So I built another personal app that solves a very specific need: wanting to learn about things via audio whilst out and about.

The result was Orra – a personal podcast / audio narration tool. I give it a topic and a target duration and it hands back an MP3.

How it works

It’s a Flask app: I give it a description of what I want to learn about and a target duration, that gets sent to an OpenAI LLM with a system prompt telling it to write me a transcript, and an OpenAI text-to-speech model turns that into an audio file.

I also played around with ElevenLabs, and the quality is genuinely impressive. I stuck with OpenAI’s model in the end because it’s significantly cheaper and plenty good enough for my needs — it doesn’t need to be VO quality.

A 15 minute episode costs a few dollars, so it’s not free, but it’s worth it and I make a few a month.

The app has a SQLite DB on a Hetzner box that gets wiped on every deploy. That’s fine — I listen once and download the MP3 if I want to keep it, and wiping the DB saves me building any UI for managing a growing pile of episodes.

What’s next

I’m curious about doing more in the AI / audio space. I’ve had an idea kicking around for interactive audiobooks — speak to characters, shape the story, the way old adventure books let you jump to different pages.

ElevenLabs is probably powerful enough to pull that off. I’m also curious about Suno — I haven’t written any music in years, but it’s the kind of thing that could drag me back into a studio (or at least back onto the laptop).