Home

Alexandria Audiobook Generator

Transform any book or novel into a fully-voiced audiobook using AI-powered script annotation and text-to-speech.

Quick Navigation

• Voice Types — Custom, Clone, LoRA, Voice Design — when to use each
• Voice Reference — Vocal direction lexicon and VoiceDesign prompt engineering findings
• Training Guide — Create custom voice identities with LoRA fine-tuning
• Dataset Builder — Build training datasets interactively with per-sample preview
• Batch Generation — How batching works, performance tuning, benchmarks
• Script Generation — LLM annotation pipeline, prompts, review pass
• Editor & Export — Chunk editing, render modes, Audacity export
• API Reference — REST endpoints with curl, Python, and JavaScript examples
• Troubleshooting — Common issues and solutions

Overview

Alexandria uses a multi-stage pipeline:

1. Upload a book (.txt or .md)
2. LLM annotates the text into a structured script with speakers, dialogue, and TTS directions
3. Configure voices for each character (5 voice type options + built-in LoRA presets)
4. Generate audio using Qwen3-TTS (batched for speed)
5. Edit and refine individual lines in the web editor
6. Export as a single MP3 audiobook or per-speaker Audacity tracks

Voice System

Alexandria supports 5 voice types, each suited to different needs:

Type	Model	Best For	Batched?
Custom	CustomVoice	Main characters (9 built-in voices with instruct control)	Yes
Clone	Base	Characters needing a specific voice (from reference audio)	Yes
LoRA	Base + adapter	Recurring characters needing a unique, persistent voice	Yes
Voice Design	VoiceDesign	Minor/throwaway characters (voice from text description)	No (sequential)
Saved Design	Base (clone)	Reusable designed voices assigned as clone references	Yes

See Voice Types for detailed guidance on choosing and configuring each type.

Requirements

• Pinokio
• An LLM server: LM Studio, Ollama, OpenAI API, or any OpenAI-compatible API
• GPU with 16+ GB VRAM recommended (NVIDIA or AMD)

No external TTS server required — Alexandria includes a built-in Qwen3-TTS engine. Model weights download automatically on first use (~3.5 GB).

Installation

1. Install Pinokio
2. Click Download and paste: https://github.com/Finrandojin/alexandria-audiobook
3. Click Install to set up dependencies
4. Click Start to launch the web interface

API Test Suite

Alexandria includes an automated test script that verifies all major API endpoints:

cd app
python test_api.py              # Quick tests (~37) — no TTS/LLM needed
python test_api.py --full       # Full tests (~49) — requires running TTS + LLM
python test_api.py --url URL    # Custom server URL (default: http://127.0.0.1:4200)

Quick mode tests config round-trips, upload, scripts CRUD, voice config, chunks, status polling, voice design listing, LoRA models/datasets listing, dataset builder CRUD, and error handling — all without loading TTS models. Full mode adds script generation, audio generation, batch rendering, voice design preview, and LoRA testing. Useful for verifying the installation works or checking nothing broke after code changes.