DrewThomasson/ebook2audiobook
DrewThomasson/ebook2audiobook · github ↗ · pipeline
◑ WATCH category dev-tooling maturity mature license Apache-2.0 apply not-now seen 2026-08-20 → 2026-08-20 #tts#voice-cloning#audiobook#multilingual#python#audio-conversion

DrewThomasson/ebook2audiobook

Multi-engine TTS converter from eBooks to audiobooks with voice cloning and 1158 languages.

What it does

  • Multi-engine TTS — XTTSv2, Bark, Fairseq, VITS, Tacotron2, Tortoise, GlowTTS, YourTTS; user picks the engine and voice.
  • Format versatility — reads 20+ ebook formats (EPUB, MOBI, PDF, TXT, HTML, RTF, DOCX, etc.); outputs M4B, MP3, FLAC, WAV, WebM, OGG, AAC.
  • Voice cloning — optional speaker voice synthesis from user audio files.
  • Language support — 1158 languages via Meta MMS TTS; includes Vietnamese.
  • SML tags — inline markup for breaks, pauses, voice switching within ebook text.

How it works

Python wrapper around multiple TTS engines (primary: XTTS v2 from Coqui), with an ebook extraction pipeline (OCR for image-heavy PDFs, chapter detection for EPUB/MOBI) and audio assembly (chapters + metadata → final audiobook file). Provides both Gradio web GUI and CLI. Runs locally (2GB RAM minimum) or remotely via Hugging Face Spaces / Google Colab / Kaggle.

Standing & signals

19.9k GitHub stars, last pushed 2026-08-09. Active development with multi-platform support (Mac, Linux, Windows, Docker), community Discord, and remote deployment options. Mature and well-maintained project.

Q2 — Apply where

not-now. Auto Social Posting's reel pipeline already uses edge-TTS (measured VI winner 2026-07-31) + whisper for word-alignment. Voice cloning via gem-boy audio is prioritized within the current stack; adopting ebook2audiobook would add heavyweight dependencies for overlapping functionality.

Q3 — Fit for current projects + sketch

Auto Social Posting could theoretically use ebook2audiobook's voice cloning (XTTS v2) to replace or complement edge-TTS, or generate longer-form audio narration if ClientCo expands into audio products. However, the current pipeline is lightweight and proven ($0.02/min VI), and ebook2audiobook's design (ebook-specific OCR, chapter handling) is orthogonal to short-form reel generation. Future fit improves only if audio scope broadens.

Q4 — Future-watch verdict

WATCH. Mature, well-maintained tool with overlapping TTS and voice cloning capabilities. Revisit if Auto Social Posting hits latency/cost limits with current engines, or if ClientCo reports need audio narration as a premium feature.