Open source / AGPL-3.0 Coming soon

txtCaster

A local-first book-to-audio studio. Point it at a PDF or EPUB, pick a voice, and it produces a real audiobook on your own machine. No cloud, no account, no telemetry.

The txtCaster library showing a finished audiobook of Alice's Adventures in Wonderland playing in the built-in mini player
The library, with a finished cast playing in the built-in player.

Two ways to listen

Narrator mode

Reads the text faithfully, word for word, in the voice you choose.

Teacher mode

Uses a local language model to re-explain the material as a lecture in its own words, turning a dense textbook chapter into something you can absorb on a walk.


What it does

Scope and control

  • Import PDF and EPUB books into a local library
  • Convert selected chapters or exact page ranges, not the whole book
  • Pick and preview a voice before committing to a long conversion

Output

  • Real MP3 files with chapter markers and EBU R128 loudness normalization
  • Spoken figure and chart descriptions with an optional local vision model
  • Built-in player with seeking, skip, adjustable speed, and a mini player

Long jobs

  • Pause and resume: conversions checkpoint as they go
  • Hardware profiling with up-front time estimates and model fit indicators
  • Overnight queue that keeps the machine awake until the job is done

Models

  • Nothing bundled: you choose what to install, from a 60 MB voice up
  • Resumable downloads, each verified by SHA256 checksum
  • Non-commercial model licenses clearly marked, never selected by default

Private by architecture

The conversion pipeline is fully offline. Documents are processed on your machine and never uploaded. There is no telemetry, no analytics, and no account system. The application touches the network only for model downloads and catalog refreshes that you initiate, and every download is verified by SHA256 checksum before use.


On the roadmap

A follow-along reading view with synchronized word highlighting, voice cloning, macOS support, and an interactive voice tutor: an explicitly opt-in mode where you talk to your book, ask questions mid-listen, jump to a chapter by voice, or have a section summarized, using your own API keys or running fully locally. Today txtCaster is an audiobook producer and an audio player; the project says plainly what it is not yet.