Audiblez Tutorial: Turn Any EPUB Into an Audiobook Locally (Free)

Audiblez is a free, MIT-licensed Python tool that turns any EPUB into an .m4b audiobook on your own machine using the Kokoro-82M speech model. Install ffmpeg and espeak-ng, run pip install audiblez, then audiblez book.epub -v af_sky. A T4 GPU converts a ~160,000-character book in about 5 minutes; an M2 CPU takes about an hour.

That's the whole pitch. No subscription, no per-book credits, no upload to anyone's server. This tutorial walks through the install, your first conversion, picking from the 20 American-English voices (and 8 other languages), what to expect from GPU vs CPU, and the one honest question everyone skips: what you're actually allowed to do with the output.

If you're weighing Audiblez against other approaches — Piper pipelines, XTTS, DIY scripts — our local audiobook generator guide covers the whole landscape. This page is the hands-on tutorial for the tool that currently wins on convenience.

---

What is Audiblez?

Audiblez (by santinic on GitHub) is an open-source command-line tool that does one job: EPUB in, .m4b audiobook out. Under the hood it feeds your book's text through Kokoro-82M, an 82-million-parameter open-weight TTS model, generates a WAV file per chapter, and stitches the per-chapter WAVs into a single .m4b file — the audiobook format your podcast or audiobook app already understands.

The quick facts:

Why the sudden interest? Audiblez had a viral moment in mid-2026 when X posts along the lines of "Audible is cooked after this" made the rounds. That framing is social-media hype — a robot narrator is not an Audible-grade human performance, and we'll be straight about that below. But the underlying tool is real, actively maintained, and genuinely useful: for the enormous pile of books that will never get a commercial audiobook, a free local conversion beats nothing by a mile.

Audiblez doesn't run the TTS model from scratch, by the way — if you want to understand (or directly script) the model doing the actual talking, see our Kokoro TTS local setup guide.

---

Step 1: Install Audiblez

Two system dependencies, one pip package. That's the entire install.

1. System dependencies. Audiblez needs ffmpeg (to assemble the .m4b) and espeak-ng (phonemization for Kokoro).

On Ubuntu/Debian:

sudo apt install ffmpeg espeak-ng

On macOS with Homebrew:

brew install ffmpeg espeak-ng

On Windows you need the same two packages installed and on your PATH; the repo documents dedicated step-by-step Windows install instructions (a virtual-environment setup), so follow those Windows steps in the Audiblez repo.

2. The Python package:

pip install audiblez

That pulls in Audiblez and its Python dependencies. Kokoro's model weights are small — the model is only 82M parameters — so this is a light install by local-AI standards. No CUDA toolkit wrestling required unless you want GPU speed (covered below).

---

Step 2: Convert Your First Book

Point Audiblez at an EPUB and give it a voice:

audiblez book.epub -v af_sky

What happens next: Audiblez parses the EPUB's chapters, synthesizes each one to a WAV file, then assembles the per-chapter WAVs into a final .m4b audiobook. When it finishes, drop the .m4b into any audiobook player and start listening.

Useful flags for real books:

• --pick — choose which chapters to convert. Handy for skipping front matter, indexes, and the 40 pages of endnotes nobody wants narrated.
• -o FOLDER — write output to a specific folder instead of the current directory.
• -s 1.5 — set speaking speed, anywhere from 0.5 to 2.0. Default pacing is measured; many listeners prefer 1.2-1.5 for nonfiction.
• --cuda — use your NVIDIA GPU. This is the difference between minutes and an hour per book (timing below).

So a realistic command for a nonfiction book on a GPU machine looks like:

audiblez book.epub -v am_michael -s 1.3 --cuda -o audiobooks/

One practical note: Audiblez takes EPUB files. If your book is a PDF or MOBI, convert it to EPUB first (Calibre handles this well) — and know that a clean EPUB with proper chapter structure produces a much better audiobook than a scanned-PDF conversion ever will.

---

Choosing Voices

Audiblez inherits Kokoro-82M's voice catalog: 20 American-English voices plus voices across 9 languages in total. Voice names encode language and gender in the prefix — af_ is American female, am_ American male, bf_ British female, and so on:

Pick with the -v flag. Two pieces of practical advice from converting real books:

1. Test on one chapter before committing. Use --pick to convert a single chapter, listen for five minutes, and only then run the full book. A voice that sounds fine in a ten-second demo can grate over eight hours.
2. Match voice to material. The male American voices tend to suit dry nonfiction; a British voice on a British novel is a small touch that makes a long listen feel less synthetic.

There's no voice cloning here — Kokoro is a fixed-voice model, so you choose from the catalog rather than feeding it a narrator sample. If voices are the part you care about most, our best local TTS models roundup compares Kokoro's catalog against the alternatives.

---

GPU vs CPU: Real Timing

This is where expectations matter. The repo's own reported numbers, using Orwell's Animal Farm (~160,000 characters) as the benchmark:

That's a clean 10x gap. Both are entirely workable — the question is workflow:

• On CPU, treat it like a batch job. Kick off a conversion before lunch or overnight. An hour per average book is fine when you're not sitting there watching it.
• On GPU, it's nearly interactive. Five minutes per book means you can convert a small library in an evening. Even a modest NVIDIA card changes the experience; and a free Colab T4 is exactly where the repo's own benchmark comes from.

Scale linearly for longer books (our estimate, not a repo number): a doorstop novel around 600,000 characters would run roughly 20 minutes on a T4 or about four hours on CPU. Animal Farm is a short book — most of your library is longer.

---

The GUI Option

Not a terminal person? Audiblez ships a graphical interface. Install the extra dependencies and launch it:

pip install audiblez pillow wxpython
audiblez-ui

On Ubuntu/Debian, GUI users also need the GTK system package the repo calls for — sudo apt install libgtk-3-dev — before audiblez-ui will run; check the repo docs for any additional GTK packages your distro needs.

Same pipeline, same voices, same output — load an EPUB, pick a voice, convert. The CLI remains the better tool for batch-converting a folder of books or running on a headless GPU box, but for a one-off conversion the GUI removes any excuse not to try it.

---

What Audiblez Can't Do

Honest expectation-setting, because the "Audible is cooked" hype deserves a reality check:

• It's not a human performance. Kokoro's narration is clean and consistent, but there's no character voicing, no dramatic pacing, no emotional read of a scene. For fiction you love, a professional narrator is still a different product.
• Unusual text trips it up. Odd names, equations, tables, code listings, and footnotes get read literally — sometimes awkwardly. Nonfiction with heavy formatting benefits from --pick to skip the worst chapters.
• EPUB only. PDFs and MOBIs need converting first, and messy source files produce messy audiobooks.
• Fixed voices. No cloning your favorite narrator. Twenty American-English voices is real variety, but it's a catalog, not a soundboard.

None of this is disqualifying. It just means Audiblez's sweet spot is the huge set of books that have no audiobook at all — technical books, older titles, niche nonfiction, public-domain classics — rather than replacing performances you can already buy.

---

The Copyright Question, Honestly

Two separate licenses are in play, and people conflate them constantly.

The tooling is clear. Audiblez is MIT-licensed. Kokoro's weights are Apache-licensed, which permits commercial use of the TTS output. Nothing about running the software is restricted.

The book is the issue. The text you feed in has its own copyright, and no software license changes that. Converting an EPUB you own into audio for your own listening is the personal-use case this tool is built around — functionally similar to having your e-reader read aloud to you. Distributing, uploading, or selling a generated audiobook of a copyrighted book is a completely different act, and the rightsholder's permission is what governs it, not Kokoro's license.

This isn't legal advice, and jurisdictions differ. The safe framing is simple: your books, your device, your ears — fine. Anyone else's ears — that's the rightsholder's call. Public-domain texts, of course, are fair game for anything.

---

Verdict: Should You Use Audiblez?

Yes — it's the easiest free EPUB-to-audiobook pipeline you can run locally right now. One pip install plus two system packages gets you from e-book to chaptered .m4b in a single command, the Kokoro voices are genuinely listenable, and the price is zero forever.

Use it if: you have a backlog of EPUBs with no audiobook edition, you're comfortable with a synthetic (but good) narrator, and you have either an NVIDIA GPU or the patience for hour-long CPU batch jobs.

Skip it if: you mostly read bestsellers that already have professional narration, or you need voice cloning and dramatic performance — that's a different toolchain, covered in our local audiobook generator guide alongside every other approach worth considering.

For most people the honest summary is: five minutes of setup, one command per book, and a growing shelf of audiobooks that simply didn't exist before. That's a very good trade for $0.

---

Sources

• Audiblez on GitHub (santinic/audiblez) — official repo: install commands, CLI flags, voice list, and the T4/M2 performance numbers cited here
• Kokoro TTS Local Setup — our deep dive on the Kokoro-82M model Audiblez runs on
• Local Audiobook Generator Guide — the broader landscape of local e-book-to-audio tools

---

FAQ