Text to Audiobook: free AI voice, MP3 or M4B

Where it runs

Works in a current Chrome, Edge, Firefox or Safari, on Windows, Mac, Linux and Chromebook computers, Android phones, iPhones and iPads. Nothing to install, no account.

The first narration downloads the voice model once: about 93 MB on a processor, or 327 MB with a graphics card (WebGPU, in current Chrome and Edge), which is several times faster.

The first file loads the audio engine (ffmpeg, about 31 MB) once, and your browser keeps it. Very long books are easiest on a computer, which has more memory than a phone.

4 simple steps.

Paste or drop text

Paste a story or a chapter, or drop a TXT, Markdown, DOCX or EPUB file. “Chapter 2” lines and # headings split it into chapters.

Pick a voice

Ten natural voices, American and British, at the speed you like. Hear a paragraph of your own text before you start.

Narrate

The AI voice reads it in your browser. Finished chapters can be played straight away.

Download

An M4B audiobook with chapters, or a ZIP of MP3s, one per chapter.

How it works

Your text is split into chapters at “Chapter” lines and headings, tidied for speech with the same rules as the script cleaner, and turned into phonemes with open pronunciation dictionaries (about 180,000 words, plus rules for names, numbers and compounds). Kokoro-82M, an Apache-licensed voice model, reads it sentence by sentence in a background worker. The audio is levelled, encoded and saved in your browser as it goes, then joined into one M4B with chapters or a ZIP of MP3s.

Have an EPUB? The EPUB to audiobook tool keeps its cover and chapter list. A PDF? PDF to audiobook takes out its page numbers, headers and footnotes first. Just a paragraph or a line, to try every voice, blend two, or fix a pronunciation? That’s Kokoro TTS.

Questions, answered.

Is my text sent to a server?

No. The voice model (Kokoro, open source) runs in your browser. It downloads once and is cached; your text and the audio stay on your device.

Is there a length limit?

No word limit and no account. Long texts take time on slower devices: with WebGPU narration is usually faster than real time, and on a processor alone about two to three times the audio’s length. Progress is saved in your browser, so you can pause and come back.

Can I use the audio commercially?

The Kokoro voices are released under the Apache 2.0 licence, which allows commercial use. You need the rights to the text, and stores have their own rules for AI narration: check them before you publish.

How do I fix a word it says wrong?

Spell it the way it sounds in your text (“Siobhan” → “Shivawn”) before narrating. Names it has never seen are read with English spelling rules.

Can it read numbers, dates and abbreviations?

Yes. Text is tidied for speech first: “Dr.” becomes “Doctor”, “$4.50” becomes “four dollars and fifty cents”, “1984” is read as a year, and footnote markers and web addresses are dropped. The script cleaner shows exactly what changes.