I have an idea for a fully automated podcast. I’m trying to get into podcasting and realizing it requires a lot of work. I need to schedule time, then edit, which takes about 5 hours to edit a 1-hour podcast.
What if I write a program that generates a podcast based on my microblog messages? On average, I post about 10 messages on my blog each week. I could automatically grab them from the Redis database using the Redis CLI to fetch messages from the past week. Then, I could use the OpenAI API to ask GPT-4 to generate a transcript.
After that, I can use OpenAI’s TTS capabilities to generate the audio. For visuals, I could create a simple video with two bubbles—one representing me and the other for the co-host—bouncing in sync with the audio. I’ll use FFmpeg to generate the video. GPT-4 can also generate the title and show notes.
The entire process could run on a cron job scheduled to execute automatically at night between Tuesday and Wednesday, uploading directly to YouTube. The only manual step would be reviewing the podcast and marking it public.