alexandria-audiobook

Why we included this project

Turning a book into a narrated audiobook usually means wiring together several TTS engines and fixing the same mistakes over and over. This project collapses that into a single pipeline: the LLM parses your text into speakers and dialogue with per-line direction hints, then a built-in Qwen3-TTS engine renders each line locally. You get voice cloning from a short reference clip, a text-driven voice designer, and optional LoRA fine-tuning if you need a character voice to stay consistent across twenty chapters. The browser-based editor lets you correct one line and re-render only that chunk, which spares you from regenerating an hour of audio after a small misattribution. Exports cover the formats listeners actually use: single MP3, chaptered M4B for Audiobookshelf and Apple Books, or per-speaker WAV tracks for Audacity. Writers, indie publishers, and hobby narrators with a capable NVIDIA or AMD GPU will find it the most practical.