Most text-to-speech tools make you trade reach for quality: commercial voices sound great but cover a dozen languages, while open models cover more but need glue code to handle a real book. The interesting move here is treating the TTS engine as a swappable backend — eight of them, from XTTSv2 and Bark to Meta's Fairseq/MMS — so one epub can be narrated in a cloned voice or in any of 1,158 languages without changing the workflow.
What Sets It Apart
- Engine choice is a feature, not a fork: XTTSv2, Bark, Fairseq, VITS, Tortoise, Tacotron2, GlowTTS, and YourTTS sit behind one interface, so you match voice quality, speed, and language to the book instead of bending the book to the tool.
- Zero-shot voice cloning from a short reference clip, with optional XTTSv2 fine-tuning when you want one consistent narrator across a whole series.
- Real book plumbing, not just a TTS demo: it ingests 20+ formats (epub, mobi, azw3, pdf, docx, even OCR'd images) and emits chapter-aware m4b/mp3/flac — the part most "read this text aloud" scripts skip.
- Runs almost anywhere: a 2 GB RAM / 1 GB VRAM floor, CPU or CUDA/ROCm/MPS/XPU/Jetson, driven from a Gradio web UI or a headless CLI in Docker/Podman.
Who It's For
Great fit if you want self-hosted, offline audiobook generation — especially in an uncommon language — or a consistent cloned narrator, and you don't mind tuning engines and settings. Look elsewhere if you expect one-click, broadcast-grade narration with zero setup: the eight-engine flexibility means quality varies by engine and language, and the best results take experimentation, GPU time, and patience with large model downloads.
