AutoAudioBook

AutoAudioBook ingests PDF or DOCX books, extracts the text, creates a reviewable annotation artifact with Gemini inline TTS tags, previews chunking, and can generate per-chunk WAV audio with Gemini 3.1 Flash TTS. I run this on a Ubuntu 24.04 VM running on two cores and 4GB ram om a Beelink ME Pro.

This project was developed in VS Code with GitHub Copilot using GPT-5.4.

Install on Ubuntu 24.04

1. Clone the repository:

git clone https://github.com/tronba/AutoAudioBook.git
cd AutoAudioBook

2. Run the installer:

sudo bash install_ubuntu_24.sh

3. Open the app in a browser:

http://<server-ip>:8000

The installer will:

• install Ubuntu packages
• create or reuse .venv
• install Python dependencies
• securely prompt for the Gemini API key
• save the key to /etc/autoaudiobook/autoaudiobook.env with restricted permissions
• install and enable the systemd service

To manage the service:

sudo systemctl status autoaudiobook
sudo systemctl restart autoaudiobook
sudo journalctl -u autoaudiobook -n 100 --no-pager

What it does

• Imports PDF or DOCX books
• Generates a reviewable annotated DOCX with inline TTS tags
• Lets you upload an approved annotation for audio generation
• Previews chunking before synthesis
• Generates WAV audio with Gemini TTS

Input file notes

• DOCX input files should mark chapter headings with the Word style Header 1
• Text before the first Chapter 1 marker is included in Chapter 1 by default
• That opening text can be split out instead if you enable the separate pre-chapter text option during generation

Tag configuration

Editable inline tag vocabulary lives in tts_tags.toml.

[[expressive_tags]]
tag = "[angry]"
min_mode = "expressive"

Each tag belongs to either expressive_tags or vocalization_tags, and each entry includes a min_mode of conservative, balanced, or expressive.

Storage layout

• storage/app.db - SQLite database
• storage/uploads/ - original uploaded source files
• storage/extracted/ - normalized extracted JSON
• storage/annotated/ - draft and approved annotation DOCX files
• storage/audio/ - reserved for future chunk and chapter audio output

Gemini configuration

Set these environment variables on the server:

• GEMINI_API_KEY
• GEMINI_TEXT_MODEL optional, defaults to gemini-2.5-flash
• GEMINI_TTS_MODEL optional, defaults to gemini-3.1-flash-tts-preview