EPUB Audiobook Reader

Turn any EPUB into an audiobook with AI voice cloning. Powered by Faster Qwen3-TTS.

Features

• Voice Cloning — clone any voice from a short audio sample
• Custom Voices — use built-in Qwen3-TTS speaker voices
• Real-time Streaming — audio plays as it's generated, no waiting
• Chapter Navigation — browse and select chapters from the sidebar
• EPUB Library — books are saved locally for quick access
• Full Audiobook Export — generate and download a complete WAV file
• Voice Manager — record reference audio directly in the browser or upload files
• Dark / Light Mode — with system preference detection
• Bilingual — automatic Chinese / English language detection and chunking

Requirements

• Python 3.10+
• NVIDIA GPU with CUDA support
• ~4GB VRAM for the 0.6B model, ~8GB for the 1.7B model

Installation

git clone https://github.com/williamcotton/epub-audiobook-reader.git
cd epub-audiobook-reader

# Install PyTorch for your CUDA version first (see https://pytorch.org/get-started/locally/)
# Example for CUDA 12.8 (RTX 50-series):
uv pip install torch --index-url https://download.pytorch.org/whl/cu128

# Then install the project
uv sync

Quick Start

# Start with the default 1.7B model (auto-downloads from HuggingFace)
epub-audiobook-reader

# Use the smaller 0.6B model
epub-audiobook-reader --model Qwen/Qwen3-TTS-12Hz-0.6B-Base

# Start without preloading a model (select in the UI)
epub-audiobook-reader --no-preload

# Enable HTTPS for microphone recording support
epub-audiobook-reader --ssl

Then open http://localhost:7861 in your browser.

CLI Options

Option	Default	Description
--model	Qwen/Qwen3-TTS-12Hz-1.7B-Base	Model to preload at startup
--port	7861	Server port
--host	0.0.0.0	Server host
--no-preload	off	Skip model loading at startup
--ssl	off	Enable HTTPS with auto-generated self-signed certificate

Environment Variables

Variable	Default	Description
QWEN3_TTS_ROOT	—	Local directory containing downloaded models
PORT	7861	Server port (overridden by --port)
MODEL_CACHE_SIZE	2	Maximum number of models to keep loaded
ACTIVE_MODELS	all	Comma-separated list of models to show in the UI
ASSET_DIR	/tmp/faster-qwen3-tts-assets	Directory for downloaded reference audio assets

Available Models

Model	Type	Size	Description
Qwen/Qwen3-TTS-12Hz-0.6B-Base	Voice Clone	~0.6B	Smaller, faster model
Qwen/Qwen3-TTS-12Hz-1.7B-Base	Voice Clone	~1.7B	Higher quality
Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice	Custom Voice	~0.6B	Built-in speakers, smaller
Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice	Custom Voice	~1.7B	Built-in speakers, higher quality

How It Works

1. Upload an EPUB — the book is parsed into chapters and text segments
2. Select a voice — choose a built-in preset, upload reference audio, or record your own
3. Play — click any chapter to start streaming audio in real-time
4. Export — optionally generate a full audiobook WAV file for offline listening

License

MIT