Back to Labs
Prototype
Podcast API
Generates two-host podcast episodes from a news feed. Script writing, voice cloning, and auto-upload to Dropbox.
Why we built this
We wanted to test if AI could produce listenable podcasts from raw news. Not just text-to-speech, but natural two-host dialogue with personality, pacing, and structure.
How we built it
Four-step GPT pipeline. Clean and filter news from Dropbox. Enhance with background context. Generate two-host dialogue (Jeff and Jane with distinct voices). Expand with round-up section. Play.ht’s PlayDialog API with custom voice clones. MP3 uploads to Dropbox. Targets 5-20 minutes per episode.
What's next
Experiment to see if AI podcasts work. Turns out they’re listenable but lack spontaneity. Good for news summaries, not conversations. We learned what we needed and shelved it. Might revisit if voice synthesis gets more natural.
Built with
React
FastAPI
Python
OpenAI API
Play.ht PlayDialog
Voice Cloning
Dropbox API
