Commercial audiobook generators charge per book or per minute, and their catalog of voices is the product. An open-source pipeline flips that: you bring the book, the AI voices are free for personal use, and the result lands in your podcast app as a proper series.
Feeding a book through TTS is the easy part. What makes the result listenable is structure: chapter-sized episodes (~10 minutes) instead of one 10-hour file, a one-line AI summary on every episode so you can navigate, resume positions that survive app restarts, and delivery through standard podcast RSS so your existing app's speed controls and lock-screen support just work.
Before chapter one, the generator produces a bonus episode in which two hosts talk through what the book argues, who it's for, and what questions to keep in mind while listening. It works like a movie trailer for books — several times we've dropped a book after the intro and saved ten hours.
Generation runs about 10x faster than playback. Upload an epub from your phone and the first episode is ready within minutes; a full 200-page book completes in 2–3 hours in the background while you're already listening.
Storage: Cloudflare R2 free tier (10GB holds dozens of books, egress is free). Voices: Microsoft neural TTS, free for personal use (swappable for ElevenLabs/Azure if you need commercial rights). Optional LLM for summaries and the intro episode: pennies per book. Total: approximately $0/month.
epub, txt, and html directly; mobi/azw3 convert to epub with one click in Calibre.
Yes — there's a whole-book translation command, e.g. listen to an English book as a Chinese audiobook. Voice selection follows the output language automatically.
Paid apps rent you the pipeline; this one is AGPL open source and runs on your own accounts, so there's no subscription and no service that can shut down and take your library with it.