Alexandria Audiobook
An open-source AI audiobook generator that runs Qwen3-TTS locally, connects to Ollama or OpenAI for script annotation, and exports MP3 or chaptered M4B.
Alexandria Audiobook is a local AI audiobook generator for people who want separate narrator and character voices, with control over individual lines. It accepts EPUB, text and Markdown books, uses an LLM to identify speakers and delivery directions, and generates speech with a built-in Qwen3-TTS engine. It's open source under the MIT license.
The browser editor lets you correct speaker assignments, change text, preview audio and regenerate selected passages. An optional LLM review checks annotation errors. Character voices can come from presets, short reference recordings or written descriptions; automatic persona generation can assign voices based on the script. LoRA training creates reusable voice identities that follow delivery instructions, while an included dataset editor helps prepare voice samples. Cloned voices don't follow those instructions.
Script annotation requires a separate LLM through an OpenAI-compatible API, including LM Studio, Ollama or OpenAI. With local backends, the book text and speech processing stay on your hardware; cloud annotation sends the book text to the chosen provider. Speech can also run on a remote Qwen3-TTS server. Model downloads need an internet connection.
The app runs on Windows, Linux and macOS, with NVIDIA GPU support on Windows and Linux and AMD GPU support on Linux. Macs, including Apple Silicon, use the slower CPU mode, as does AMD hardware on Windows. GPU use needs at least 8 GB VRAM; 16 GB RAM is recommended. Docker supports NVIDIA server deployments. Exports include a combined MP3, chaptered M4B for audiobook players, and separate speaker tracks and labels for Audacity.
