audiobook

name	audiobook
description	Convert documents (PDF, Text, Markdown, Docx) into high-quality narrated audiobooks with automatic chapter detection. Generates one MP3 file per chapter with metadata.
version	1.0.0

Audiobook

Note: Review PROFILE.md in this skill folder for user-specific narration preferences, voice settings, and chapter handling guidelines.

Master Briefing: Global brand voice at ~/.superskills/master-briefing.yaml applies automatically. Skill profile overrides when conflicts exist.

Convert documents into professional audiobooks with automatic chapter detection, generating high-quality narrated MP3 files optimized for long-form listening.

Supported Formats

• PDF (.pdf) - Text extraction from PDF documents
• Plain Text (.txt) - Plain text files
• Markdown (.md, .markdown) - Markdown documents
• Word Documents (.docx) - Microsoft Word files

Tools

DocumentParser.py (in src/):

• Multi-format document text extraction
• UTF-8 and fallback encoding support
• PDF, Docx, Markdown, and plain text parsing
• Validation and error handling

ChapterDetector.py (in src/):

• Automatic chapter boundary detection
• Multiple detection strategies (markdown, numbered, pagebreak, custom)
• Smart fallback for documents without clear chapters
• Chapter splitting for oversized content
• Configurable custom regex patterns

AudiobookConfig.py (in src/):

• Voice profile configuration management
• Singleton pattern for efficient config loading
• Validation of voice settings
• Environment variable fallback support

AudiobookGenerator.py (in src/):

• Complete audiobook generation orchestration
• Chapter-by-chapter audio generation
• ElevenLabs API integration with model fallback
• Metadata JSON generation
• Progress tracking and logging

Core Workflow

1. Document Parsing

• Load document from supported format (PDF, TXT, MD, DOCX)
• Extract full text content
• Handle encoding and format-specific quirks
• Validate text extraction success

2. Chapter Detection

• Analyze text structure using selected strategy
• Detect chapter boundaries automatically
• Extract chapter titles and content
• Handle edge cases (no chapters, oversized chapters)

3. Audio Generation

• Optimize text for natural speech delivery
• Generate audio for each chapter using ElevenLabs
• Apply audiobook-specific voice settings
• Calculate duration and WPM metadata
• Save individual chapter MP3 files

4. Metadata Export

• Create comprehensive metadata JSON
• Include chapter titles, durations, word counts
• Track generation timestamp and settings
• Provide formatted duration strings

Chapter Detection Strategies

auto (default):

• Try detection strategies in priority order
• Prefer markdown headings → numbered patterns → page breaks
• Fallback to single chapter or size-based splitting

markdown:

• Detect chapters from # or ## headings
• Best for markdown-formatted documents
• Preserves heading text as chapter titles

numbered:

• Detect "Chapter 1", "CHAPTER 1", "1. Title" patterns
• Case-insensitive matching
• Extracts chapter titles when present

pagebreak:

• Split on form feed (\f) or triple-dash (---) dividers
• Attempts to extract titles from first lines
• Falls back to "Section N" titles

custom:

• Use custom regex patterns from config or env var
• Flexible for non-standard document structures
• Pattern errors logged but don't crash generation

Chapter Announcements

Every chapter audio file begins with a spoken announcement that includes:

For regular chapters:

• Book title (extracted from filename)
• Chapter number
• Chapter title (if available)

For special sections (Prologue, Epilogue, Foreword, etc.):

• Book title
• Section name and subtitle (if available)

Example announcements:

• Regular: "The Great Novel. Chapter 1. The Beginning."
• Special: "The Great Novel. Prologue: The Dark Past."

This feature ensures listeners always know where they are in the audiobook, especially useful when:

• Resuming playback after a break
• Navigating between chapter files
• Listening to audiobooks split across multiple sessions

Implementation: The announcement is prepended to the chapter text before sending to the text-to-speech API, creating a natural pause between the announcement and the chapter content.

Prologue and Epilogue Support

The skill automatically detects and processes special book sections alongside regular chapters:

Supported special sections:

• Prologue / PROLOGUE: Story introduction or background
• Epilogue / EPILOGUE: Story conclusion or aftermath
• Foreword / FOREWORD: Author or expert introduction
• Afterword / AFTERWORD: Author reflections or commentary
• Preface / PREFACE: Book context or purpose

Detection patterns:

• Sections must appear on their own line (heading style)
• Case-insensitive matching (e.g., "Prologue", "PROLOGUE", "prologue")
• Optional subtitle after colon: Prologue: The Dark Past
• Properly ordered in output: Prologue → Chapters → Epilogue

Example document structure:

Prologue: The Dark Past

[Prologue content...]

Chapter 1: The Beginning

[Chapter 1 content...]

Epilogue: What Happened Next

[Epilogue content...]

This generates 3 MP3 files:

• 001_prologue_the_dark_past.mp3
• 002_chapter_1_the_beginning.mp3
• 003_epilogue_what_happened_next.mp3

Text Cleaning

The audiobook skill automatically cleans extracted PDF text to improve narration quality:

What Gets Removed

• Page numbers: "42", "Page 42", "42 | Chapter Title", "Chapter Title | 42", "42 of 200"
• Headers/footers: Repeated chapter titles, copyright notices (©), date stamps, "All rights reserved"
• Hyphenated words: Word breaks at line endings are rejoined ("exam-\nple" → "example")
• Excessive whitespace: Multiple blank lines reduced to maximum 2, multiple spaces to 1
• Duplicate lines: Repeated headers or OCR errors removed

Configuration

Text cleaning is enabled by default for PDF files. To disable:

Python API:

generator = AudiobookGenerator(clean_text=False)

Custom Patterns:

from superskills.audiobook.src.TextCleaner import TextCleaner

cleaner = TextCleaner(
    remove_page_numbers=True,
    remove_headers_footers=True,
    custom_patterns=[
        r'^Confidential$',      # Remove "Confidential" headers
        r'^\d{4}-\d{2}-\d{2}'  # Remove date stamps
    ]
)

cleaned_text = cleaner.clean(raw_text)

Why Text Cleaning Matters:

• Natural flow: Narrator shouldn't read "Page 42" between paragraphs
• Professional quality: Copyright notices and headers disrupt listening experience
• Correct pronunciation: Rejoined hyphenated words prevent awkward pauses
• Better chapter detection: Cleaned text produces more accurate chapter boundaries

Quality Checklist

• Document format supported and parsed successfully
• Chapters detected accurately (verify count and titles)
• Audio quality high (no artifacts or pronunciation issues)
• Chapter MP3 files generated with correct naming
• Metadata JSON complete and accurate
• Total duration and word counts match expectations
• Voice settings appropriate for long-form narration
• No page numbers or headers in generated audio (listen to sample)

Usage

Python API

Basic Usage:

from superskills.audiobook.src import AudiobookGenerator

generator = AudiobookGenerator(output_dir="audiobooks")
result = generator.generate_audiobook("my_book.pdf", output_prefix="my_book")

print(f"Generated {result.total_chapters} chapters")
print(f"Total duration: {result.total_duration_seconds / 3600:.1f} hours")
print(f"Metadata: {result.metadata_file}")

With Custom Settings:

generator = AudiobookGenerator(
    output_dir="audiobooks",
    profile_type="audiobook",
    chapter_strategy="markdown",
    custom_chapter_patterns=[r"^PART \d+:"]
)

result = generator.generate_audiobook(
    file_path="novel.docx",
    output_prefix="my_novel"
)

for chapter_file in result.chapter_files:
    print(f"Generated: {chapter_file}")

Access Individual Components:

from superskills.audiobook.src import DocumentParser, ChapterDetector

# Parse document
parser = DocumentParser()
text = parser.parse_file("book.pdf")

# Detect chapters
detector = ChapterDetector()
chapters = detector.detect_chapters(text, strategy="auto")

for chapter in chapters:
    print(f"Chapter {chapter.number}: {chapter.title} ({chapter.word_count} words)")

CLI Usage

Basic Generation:

superskills call audiobook --input book.pdf

With Custom Prefix:

superskills call audiobook --input novel.docx --output-prefix "my_novel"

With Chapter Strategy:

superskills call audiobook --input memoir.txt --chapter-strategy markdown

Process Specific Chapters:

# Process only first chapter (for testing)
superskills call audiobook --input book.pdf --chapter-range 1

# Process chapters 1-3
superskills call audiobook --input book.pdf --chapter-range 1-3

# Process chapters 5-10
superskills call audiobook --input book.pdf --chapter-range 5-10

Use cases for chapter-range:

• Testing: Quickly test voice/settings on first chapter before full conversion
• Partial conversion: Generate only specific sections of interest
• Resource management: Process large books in batches to manage costs/time
• Corrections: Re-generate specific chapters after source edits

Notes:

• Chapter numbers are 1-indexed (first chapter is 1, not 0)
• Range is inclusive: 1-3 processes chapters 1, 2, and 3
• If end chapter exceeds total chapters, processes up to last available chapter
• If start chapter exceeds total chapters, raises an error

Custom Voice Profile:

superskills call audiobook --input textbook.pdf --voice-profile narration

Configuration

Voice Profiles (voice_profiles.json)

Create voice_profiles.json from template:

cp voice_profiles.json.template voice_profiles.json

Edit with your ElevenLabs voice settings:

{
  "audiobook": {
    "voice_id": "your_voice_id_here",
    "voice_name": "Your Audiobook Voice",
    "language": "English",
    "model": "eleven_turbo_v2_5",
    "speed": 0.95,
    "stability": 0.65,
    "similarity_boost": 0.85,
    "style": 0.20
  }
}

Environment Variables (.env)

# Required
ELEVENLABS_API_KEY=your_api_key_here
ELEVENLABS_VOICE_ID=your_voice_id_here

# Optional
AUDIOBOOK_CHAPTER_STRATEGY=auto
AUDIOBOOK_CUSTOM_CHAPTER_PATTERN=^Chapter \d+:

Voice Settings Explained

speed (0.7-1.2): Narration speed

• 0.95 recommended for audiobooks (slightly slower for comprehension)

stability (0.0-1.0): Voice consistency

• 0.65 recommended (balanced between consistent and natural)

similarity_boost (0.0-1.0): Voice similarity to original

• 0.85 recommended (natural but clear)

style (0.0-1.0): Expressiveness variation

• 0.20 recommended (minimal variation for long-form)

Output

Chapter Files

• Naming: {prefix}_chapter_{number:02d}_{title}.mp3
• Example: my_book_chapter_01_introduction.mp3
• Format: MP3 with 320kbps quality
• One file per chapter

Metadata File

• Naming: {prefix}_metadata.json
• Example: my_book_metadata.json

Structure:

{
  "title": "my_book",
  "source_file": "my_book.pdf",
  "total_chapters": 15,
  "total_duration_seconds": 32400.5,
  "total_duration_formatted": "9h 0m 0s",
  "total_word_count": 85000,
  "average_wpm": 157,
  "generated_at": "2026-01-09T14:30:00",
  "profile_type": "audiobook",
  "voice_name": "Your Voice",
  "chapters": [
    {
      "number": 1,
      "title": "Introduction",
      "file": "my_book_chapter_01_introduction.mp3",
      "duration_seconds": 420.5,
      "duration_formatted": "7m 0s",
      "word_count": 1100,
      "wpm": 157
    }
  ]
}

Avoid

• Poor Chapter Detection: Manual chapter review → Verify chapter strategy matches document structure
• Wrong Voice Settings: Generic settings → Use audiobook-optimized profile (slower speed, balanced stability)
• Large Files Without Splitting: Processing entire book at once → Auto-splits oversized chapters for manageable API calls
• Missing Metadata: Generate audio only → Always create metadata JSON for tracking and organization
• Ignoring Format Issues: Assume parsing worked → Validate text extraction and chapter detection before generating audio

Escalate When

• Document exceeds 200,000 words (cost and time concerns)
• PDF extraction quality poor (scanned images, complex layouts)
• Chapter detection completely fails (needs manual chapter markers)
• API rate limits hit during generation (needs throttling or batch processing)
• Voice compatibility issues persist across all model fallbacks
• Custom chapter patterns needed but unclear from document structure
• Multi-language content requires different voice profiles per section

Integration Examples

With Author Skill (Script → Audiobook):

from superskills.author.src import Author
from superskills.audiobook.src import AudiobookGenerator

# Generate book content
author = Author()
book_content = author.write_long_form("Write a short story about...")

# Save to file
with open("story.txt", "w") as f:
    f.write(book_content)

# Convert to audiobook
generator = AudiobookGenerator()
result = generator.generate_audiobook("story.txt", "my_story")

With Publisher (Audiobook Distribution):

from superskills.audiobook.src import AudiobookGenerator
from superskills.publisher.src import Publisher

# Generate audiobook
generator = AudiobookGenerator()
result = generator.generate_audiobook("book.pdf", "audiobook")

# Publish chapter files
publisher = Publisher()
for chapter_file in result.chapter_files:
    publisher.upload_to_platform(chapter_file, platform="podcast")

Technical Notes

• API Costs: ElevenLabs charges per character. 100,000 words ≈ 500,000 characters.
• Generation Time: ~1 minute per 5,000 words (varies by API load)
• Model Fallback: Automatically tries turbo → monolingual → flash if voice incompatible
• Memory Usage: Processes chapters sequentially to avoid memory issues
• File Limits: PDF parsing limited by PyPDF2 capabilities (text-based PDFs only)