Generate Audiobooks Locally Free 2026: EPUB to Audio

To generate an audiobook locally for free in 2026, the fastest path is Audiblez, an MIT-licensed Python tool that turns an EPUB into a finished .m4b using the Kokoro-82M voice model — install it with pip install audiblez plus ffmpeg and espeak-ng, then run one command. It works fully offline on CPU, and with an NVIDIA GPU it narrates a full-length book (~160,000 characters) in roughly five minutes. If you want cloned-voice narration instead of a stock voice, use Pandrator (a GUI app that wraps XTTS v2, Chatterbox and others) or pair epub2tts with a cloning model. The one thing to get right before you publish anything: voice licensing. Kokoro (Apache 2.0) and Chatterbox (MIT) are safe for commercial audiobooks; XTTS v2 is non-commercial only.

Everything here runs on your own machine — no per-character cloud fees, no upload of your book to anyone's server, and no monthly subscription. The trade is that you supply the compute and do a little setup. Below are the three tools worth your time, what each is actually good at, and the licensing rules that decide whether you can sell the result.

What is the easiest way to generate an audiobook locally for free?

Audiblez. It is the most "one command, done" option of the three local tools covered here. You point it at an .epub file, pick a voice, and it produces a chaptered .m4b audiobook you can drop straight into VLC, Apple Books or any audiobook player. Under the hood it uses Kokoro-82M, an 82-million-parameter text-to-speech model whose v1.0 release (January 27, 2025) ships 54 voices across 8 languages and sounds far more natural than the robotic system voices most people associate with "text to speech."

Install it on any machine with Python 3, ffmpeg and espeak-ng:

pip install audiblez
# system deps (example for Debian/Ubuntu):
# sudo apt install ffmpeg espeak-ng
audiblez book.epub -v af_sky -o ./audiobook

That is the whole flow. The tool first renders each chapter to a .wav file, then muxes everything into a single .m4b with chapter markers. It runs on CPU by default and uses your NVIDIA GPU automatically if PyTorch sees CUDA. There is also a small wxPython GUI (audiblez-ui) if you would rather click than type. The project is open source under the MIT license, which is one reason it is safe to lean on for real work. You can read the source and full option list on the Audiblez GitHub repository.

Which local audiobook tools should you actually use in 2026?

There are dozens of half-finished scripts on GitHub. These three are the ones that are maintained, do EPUB (or PDF) to a real audiobook file, and run entirely offline:

A few honest notes on the table. Audiblez and epub2tts-kokoro both use Kokoro, which does not do voice cloning — you choose from its built-in voices, full stop. They differ mostly in ergonomics: Audiblez is the more polished "drop in a book" experience, while epub2tts-kokoro (and its sibling epub2tts) is a leaner CLI that fits nicely into batch scripts and needs Python 3.11. Pandrator is the heavyweight — a Windows-friendly GUI with a one-click installer that can drive XTTS v2, Chatterbox, Kokoro, Silero and others, add LLM-based text cleanup, and do actual voice cloning. It is more to install, but it is the one to pick if you want a narrator that sounds like a specific voice.

How do you clone a voice for narration locally?

Cloned-voice narration means handing the tool a short reference clip and having it read your whole book in that voice. Kokoro cannot do this, so Audiblez and epub2tts-kokoro are out for cloning. Your two realistic local options are:

• Pandrator + XTTS v2 — XTTS v2 does zero-shot cloning from about six seconds of reference audio and supports many languages. Pandrator wraps it in a GUI and can even fine-tune XTTS for a better match. Great quality, but see the licensing warning below before you sell anything made with it.
• Pandrator + Chatterbox — Chatterbox is Resemble AI's open-source model (built on a 0.5B Llama backbone) with zero-shot voice cloning and emotion control. It is the cloning engine to reach for when you need a commercial-safe result, because of its license.

If you only want to learn cloning before committing to a full book, our walkthroughs on the XTTS v2 voice cloning guide and the broader local AI voice cloning options cover reference-clip prep, sample rates and the common quality pitfalls in detail.

Can you sell an audiobook made with these tools? (Licensing)

This is the part people skip and regret. Two licenses matter: the license of the app (Audiblez, Pandrator, etc.) and the license of the voice model that actually generates the audio. The model license is the one that governs whether you may sell the output. Here is the verified state in mid-2026:

The headline: Kokoro (Apache 2.0) and Chatterbox (MIT) let you sell the audiobook; Apache 2.0 weights explicitly allow commercial deployment, and Chatterbox's MIT license puts no restriction on commercial use. XTTS v2 is the trap. It is released under the Coqui Public Model License (CPML), which permits non-commercial use only — personal projects, research and hobby narration are fine, but selling the audio is not. Worse, Coqui Inc. shut down in January 2024, so there is no longer anyone to buy a commercial XTTS license from. Treat XTTS v2 output as non-commercial, period. You can confirm the terms in the XTTS-v2 model card and LICENSE.txt.

Practical rule of thumb: if you are making an audiobook to sell or publish, narrate with Kokoro (stock voice) or Chatterbox (cloned voice). Use XTTS v2 only for your own listening. None of this is legal advice — read the actual license for your exact use — but those three lines cover the cases that trip people up.

How long does it take to convert a book offline?

Speed depends almost entirely on whether you have a GPU. The numbers below combine the maintainer's published figures with our own rough timing on a single machine, so treat them as approximate and hardware-dependent, not a controlled benchmark.

In our own testing, the pattern held: Kokoro on a CPU-only laptop is comfortably real-time-plus — a couple-hundred-page novel finishes in well under an hour while you do something else — and a modest NVIDIA card collapses that to minutes. XTTS v2 and Chatterbox are heavier per second of audio than Kokoro and benefit much more from a GPU; on CPU-only they can crawl, so if cloned voices are the goal, plan on having a GPU. Kokoro's tiny 82M footprint is exactly why it is the default for "I just want the book read to me" — it runs on hardware that chokes the bigger cloning models.

What about PDF files instead of EPUB?

EPUB is the cleaner input because it already has structured chapters and clean text. PDFs are messier — columns, headers, footers and page numbers all leak into the narration if you are not careful. Of the three tools, Pandrator handles PDF directly, which is its other advantage over the Kokoro-only pair. If you only have a PDF and want to use Audiblez or epub2tts, convert it to EPUB first (Calibre is the standard, free tool for this) and skim the result to strip page furniture before narrating. Clean text in, clean audiobook out.

Key Takeaways

1. Audiblez is the fastest free local path — pip install audiblez, point it at an .epub, get a chaptered .m4b. MIT-licensed, uses Kokoro-82M, runs offline on CPU or GPU.
2. For cloned-voice narration, use Pandrator — it wraps XTTS v2, Chatterbox and Kokoro in a GUI with an installer, takes PDF or EPUB, and is the no-terminal option.
3. epub2tts-kokoro is the scriptable middle ground — a lean Kokoro CLI (Python 3.11) for batch jobs.
4. Licensing decides if you can sell it: Kokoro (Apache 2.0) and Chatterbox (MIT) are commercial-safe; XTTS v2 (CPML) is non-commercial only, and with Coqui shut down since January 2024 there is no commercial license to buy.
5. A GPU turns hours into minutes. Kokoro narrates a full book in ~5 minutes on an NVIDIA T4 vs ~1 hour on an M2 CPU; cloning models need a GPU far more than Kokoro does.

Next Steps

• Want the deepest setup walkthrough for the voice engine behind Audiblez? Read our Kokoro TTS local setup guide.
• Planning to narrate in a specific voice? Start with the XTTS v2 voice cloning guide — just remember it is non-commercial only.
• Comparing every local cloning option before you commit? See the local AI voice cloning overview.
• Building a full audio pipeline (intros, multi-voice, publishing)? Our podcast automation tutorial covers chaining TTS into a repeatable workflow.