Skip to content

Audiobook Export

Overview

The Audiobook plugin turns your chapters into spoken MP3 files using Text-to-Speech. It sits on top of the Export plugin and since v0.11 delegates all TTS work to manuscripta.

Generation runs asynchronously in the background. A progress dialog shows each chapter's status via Server-Sent Events while you keep working. Finished files are persisted under uploads/{book_id}/audiobook/ so they survive browser reloads and backend restarts.

TTS Engines

Bibliogon ships adapters for several engines with different tradeoffs:

  • Edge TTS (default): Microsoft's free cloud TTS. Large voice selection, online only, no account needed.
  • Google Cloud TTS: Standard, WaveNet, Neural2, Studio and Journey quality tiers. Requires a service account JSON which is stored Fernet-encrypted in config/google-credentials.enc. Free monthly quota, then paid.
  • ElevenLabs: High-quality, very natural voices. Paid. The API key is stored in audiobook.yaml via Settings > Audiobook, validated against the ElevenLabs API before being saved.
  • pyttsx3: Offline, uses the OS-native speech synthesis. No internet needed, lower quality.
  • Google Translate TTS (gTTS): Free fallback, minimal quality, no account needed.

You can pick the engine globally in Settings > Audiobook and override it per book in the book metadata editor.

Per-Book Configuration

As of v0.11 every audio-specific setting lives on the book itself, not as a plugin-global default. Open BookEditor > Metadata > Audiobook to configure:

  • Engine, voice, speed, merge mode, filename prefix
  • Overwrite existing files: When enabled the content-hash cache is ignored on the next export and every chapter is regenerated. Default off.
  • Skip chapter types: Checkbox list of all 26 chapter types, grouped into "Present in this book" and "Other types". Selected types are dropped from the synthesis pipeline. Marketing types (also_by_author, excerpt, call_to_action) are selected by default so audiobooks don't ship with a "buy my next book" pitch.
  • Announce chapter number: When enabled the TTS speaks an ordinal intro ("First chapter", "Second chapter") before each chapter. Default off.

All settings are persisted via the metadata editor's global Save button - there is no per-field auto-save.

Content-Hash Cache

Every generated chapter gets a sidecar .meta.json with a hash over (content, engine, voice, speed). On the next run, chapters whose hash still matches are reused from the persisted audiobook directory instead of being re-synthesised. That saves both time and API credits for paid engines. The per-book "Overwrite existing files" flag disables the cache when you explicitly want a full regeneration.

Dry Run and Cost Estimate

The dry-run mode generates a short sample MP3 from the first non-skipped chapter that has content, and returns a cost estimate for the full export in the response headers. Use it before your first expensive ElevenLabs or Google Cloud run to hear the voice and ballpark the price.

Output Modes

The merge mode controls which files get produced:

  • merged: A single MP3 for the whole book.
  • separate: One MP3 per chapter.
  • both: Both per-chapter files and a merged output.

All files can be downloaded individually or as a ZIP, and remain available after a browser reload or backend restart. If a persisted audiobook already exists, the UI warns with a confirm dialog before overwriting (unless the per-book "Overwrite existing files" flag is deliberately set).