Generate a podcast file through an asynchronous Worker API
At a glance
A blogger built podcast-it to turn article text into narration and downloadable audio using OpenAI and Cloudflare. This workflow is listed under Audio, Content Tools, APIs.
- Agents / tools
- OpenAI API, Cloudflare Workers, R2, D1, podcast-it
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Have Node.js and Cloudflare/OpenAI access. Download h93xV2/podcast-it and run `npm install`. Set OPENAI_API_KEY in local `.dev.vars`; run `npx wrangler login`. Create resources with `npx wrangler r2 bucket create podcasts`, `npx wrangler d1 create podcasts` and `npx wrangler queues create episodes`. Put the returned database ID in `wrangler.jsonc`, retaining its bucket, database and queue bindings. 2) Run `npm run dev`, which applies local migrations, and open http://localhost:8787 for the API interface. 3) POST JSON to `/api/episodes` with unique `slug`, article text as `content`, `episodeTitle`, `showTitle`, and `hosts` containing a name and voice such as alloy. The queue consumer generates a narration script and speech, storing audio in R2 and metadata in D1. 4) Poll `/api/episodes/<slug>` until status is complete and `audioFile` is populated. Download `/api/audio/<audioFile>` and listen.
How it went
The tutorial demonstrates a 202 creation response, pending-to-complete status and audio retrieval. Duplicate slugs return 409.
Worth knowing
The linked repository adds a queue and title fields to the article's example. Its package script is npm run dev; the README's npm start is stale.