kokoro-audiobook

Convert a plain-text book into an MP3 audiobook using Kokoro text-to-speech.

GPU-accelerated, fully resumable, with a small web UI for control.

Features

• 🎙️ Natural speech via Kokoro-82M (kokoro>=0.9.4)
• ⚡ GPU synthesis (~8× realtime); also runs on CPU
• ⏯️ Resumable — interrupt any time, rerun to continue exactly where it stopped
• 🌐 Web control panel: pause / resume / stop
• 🔇 Configurable silence inserted between sentences
• 📦 ffmpeg collects all chunks into a single audiobook.mp3

Requirements

• Python 3.10+
• ffmpeg and espeak-ng — sudo apt install ffmpeg espeak-ng
• Optional: an NVIDIA GPU

Install

python3 -m venv .venv
.venv/bin/pip install -r requirements.txt

# torch — pick one:
.venv/bin/pip install torch                              # CPU / modern GPU
# Maxwell GPUs (GTX 750 Ti / 9xx) need the CUDA 12.1 build:
.venv/bin/pip install torch==2.5.1 --index-url https://download.pytorch.org/whl/cu121

Usage

Put your text in book.txt, then run the whole pipeline:

./collector.sh

It converts the entire book and produces audiobook.mp3. While it runs, open http://127.0.0.1:8765/ to watch progress and pause / resume / stop.

Converter only

.venv/bin/python book_to_speech.py            # convert (resumable)
.venv/bin/python book_to_speech.py --limit 8  # quick test: only 8 chunks

Quick demo

.venv/bin/python joke_tts.py "Your one-liner here"

How it works

1. book.txt is cleaned (front-matter and markup stripped) and split into 1–2 sentence chunks.
2. Pending chunks live in book-work.txt, one per line — it shrinks from the top as work completes, so it doubles as the resume ledger.
3. Each chunk → Kokoro → an MP3 in mp3/part_NNN/, with trailing silence.
4. collector.sh ffmpeg-concatenates every chunk into audiobook.mp3.

Crash-safe: MP3s are written atomically and the worker skips any chunk that is already done, so re-running never duplicates or loses audio.

Notes

• Text cleaning in build_clean_text() (book_to_speech.py) is tuned for one ebook export — adjust the start marker and regexes for your own source.
• Bring your own text. Do not redistribute audio generated from copyrighted books.

License

MIT — see LICENSE.