Alexandria
Finrandojin/alexandria-audiobookReviewed by AI
Checked by RepoPicks Radar on
What is Alexandria?
You give it a book as .txt, .md or .epub and it produces a narrated audiobook with a different voice for each character. An AI language model you supply splits the text into lines with speakers and tone directions, then a local speech model reads them; you can fix any line in a browser editor and export MP3, a chaptered M4B or separate tracks for Audacity. It needs an NVIDIA card with 8 GB of video memory for decent speed, about 20 GB of disk, a separate language-model server, and it installs through Pinokio or Docker. A Google Colab notebook is also offered.
Alexandria is free and open source (MIT) and can replace ElevenLabs, Murf, and Speechify.
How easy is it to install Alexandria?
D · terminal, several tools or a specific graphics card · Windows
- Windows D
- macOS D
- Linux D
Windows and Linux are the best case and still a D: the install goes through Pinokio or Docker with the NVIDIA Container Toolkit, you need an NVIDIA card with at least 8 GB of video memory and driver 550+, about 20 GB of disk, and you must separately install and start a language-model server such as LM Studio or Ollama before the first button works. On a Mac it is also a D and the README says Apple Silicon runs on processor only, "functional but slow"; AMD cards need Linux with ROCm.
Coming soon: install in 1 click with the RepoPicks app.
Alexandria on GitHub
- Stars
- 1,097on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +9stars in 6 days
Stars over time
Facts
- Runs on
- Windows, Linux, macOS
- You need to install
- Python, NVIDIA graphics card, FFmpeg, Docker
- To build from source
- Docker, NVIDIA graphics card, FFmpeg
- Disk space
- not determined
- Accounts or keys
- OpenAI or another OpenAI-compatible language model service (not needed if you run LM Studio or Ollama on your own machine), Hugging Face (HF_TOKEN) for downloading the speech models, ngrok, only for the Google Colab route
- Cost
- Free · The program is free (MIT). It can run entirely on your machine with a free local language model, but if you point it at a paid service such as the OpenAI API you pay that service for every book you process.
- License
- MIT (Permissive)
- Can replace
- ElevenLabs, Murf, Speechify, Audible
- Last change on GitHub
- Aug 2, 2026
Alexandria risk signals (5)
- Attentionvoice cloning from a short sample
The features list says "Voice Cloning - Clone any voice from a 5-15 second reference audio sample", and the Training tab fine-tunes a persistent copy of a voice from your own recordings. Copying a real person's voice without their permission can break the law in several countries and the README says nothing about consent.
AI review
- Attentionweb interface and api without a password
The web interface and the whole REST API (upload, generate, train, download) answer on port 4200 with no login shown anywhere in the README, the Dockerfile sets "ENV ALEXANDRIA_HOST=0.0.0.0" and docker-compose publishes "4200:4200". On a shared machine or a server, anyone who can reach that port can use your models and your files, and the Colab route opens an ngrok tunnel to it.
AI review
- Infoyour book text goes to the language model you choose
Step 2 says "Click Generate Annotated Script — this sends the book to your LLM", and the Setup tab takes an "LLM Base URL" and "LLM API Key". If you pick a cloud service such as the OpenAI API, the full text of the book leaves your computer; with LM Studio or Ollama it stays local.
AI review
- Inforights over the book and the result
The Script tab accepts .epub files and the pipeline is built to narrate "any book or novel". Making an audiobook of a book you did not write, and above all sharing or selling it, needs the rights holder's permission.
AI review
- Infolarge downloads on first use
The README says "The TTS models are not included in the install. They download automatically from Hugging Face the first time you generate audio", "Each model variant is ~3.5 GB" and the disk requirement is "~20 GB (8 GB venv/PyTorch, ~7 GB for model weights, working space for audio)". The interface may look frozen during that download.
AI review
Evidence is shown in its original language.
What Radar could not check about Alexandria
- Whether the Pinokio install finishes on your machine, since it builds a Python environment and PyTorch at install time.
- The speed figures in the README (3-6x real time), measured by the author on an AMD RX 7900 XTX with 24 GB.
- How good the narration and the character voices actually sound for your book and language.
- Real disk and video-memory use after the models, the LoRA adapters and the generated audio pile up.
- Whether the Google Colab notebook still works on the free tier and how long a book takes there.
- How much a cloud language model would cost to annotate a full-length book.