🎙️ Audio Content Generator

Generate high-quality audiobooks, podcasts, or educational audio content on demand using AI-written scripts and ElevenLabs text-to-speech.

Quick Start

Create an audiobook chapter:

User: "Create a 5-minute audiobook chapter about a dragon discovering friendship"

Generate a podcast:

User: "Make a 10-minute podcast about the history of coffee"

Produce educational content:

User: "Generate a 15-minute educational audio explaining how neural networks work"

Content Formats

Audiobook

Style: Narrative storytelling with emotional depth

• Clear beginning, middle, and end
• Descriptive language and vivid imagery
• Dramatic pacing with thoughtful pauses
• Emotional tone that matches the story
• Use voice effects like

  [whispers]

  ,

  [excited]

  ,

  [serious]

  for impact

Example Structure:

[Opening hook - set the scene]
[long pause]

[Story development with character emotions]
[short pause] between sentences
[long pause] between paragraphs

[Climax with dramatic tension]
[long pause]

[Resolution and emotional closure]

Podcast

Style: Conversational and engaging

• Warm, welcoming intro (15-30 seconds)
• Main content with natural flow
• Transitions between topics
• Memorable outro with key takeaways
• Conversational tone throughout

Example Structure:

**Intro:** "Welcome to [topic]. I'm excited to share..."
[short pause]

**Main Content:** "Let's start with... [topic 1]"
[long pause] between segments

**Outro:** "Thanks for listening! Remember..."

Educational Content

Style: Clear explanations for learning

• Simple introductions to complex topics
• Step-by-step breakdowns
• Real-world examples and analogies
• Recap of key concepts at the end
• Enthusiastic delivery with

  [excited]

  for important points

Example Structure:

**Introduction:** What is [topic] and why it matters?

**Main Content:**
- Concept 1: Explanation + Example
- Concept 2: Explanation + Example
- Concept 3: Explanation + Example

**Summary:** Key takeaways and next steps

Length Guidelines

Word Count to Duration Conversion:

• 5 minutes = ~375 words
• 10 minutes = ~750 words
• 15 minutes = ~1,125 words
• 20 minutes = ~1,500 words
• 30 minutes = ~2,250 words

Pacing: Average conversational speed is ~75 words per minute

Practical Limits:

• Minimum: 2 minutes (~150 words)
• Maximum: 30 minutes (~2,250 words)
• Sweet spot: 5-15 minutes for best engagement

Workflow Instructions

Step 1: Understand the Request

Parse the user's request for:

1. Content type (audiobook, podcast, educational, or inferred from topic)
2. Topic/theme (what should the content be about)
3. Target length (how many minutes)
4. Tone/style (dramatic, casual, educational, etc.)
5. Special requests (specific voice, emphasis on certain points)

Step 2: Calculate Word Count

target_words = target_minutes × 75

Example: 10 minutes = 10 × 75 = 750 words

Step 3: Generate the Script

Write the complete script following these rules:

Content Guidelines:

• Start strong with an engaging hook
• Maintain natural, conversational flow
• Use active voice and simple sentence structure
• Include relevant examples and stories
• End with a satisfying conclusion

Formatting Rules:

• Add

  [short pause]

  after sentences (use sparingly, not every sentence)
• Add

  [long pause]

  between paragraphs or major sections
• Use voice effects strategically:

  [whispers]

  ,

  [shouts]

  ,

  [excited]

  ,

  [serious]

  ,

  [sarcastic]

  ,

  [sings]

  ,

  [laughs]

• Write numbers as words: "twenty-three" not "23"
• Spell out acronyms first time: "AI, or artificial intelligence"
• Avoid complex punctuation (em-dashes work, but semicolons don't read well)
• Remove markdown formatting before TTS conversion

Step 4: Present the Script

Show the script to the user and ask:

Here's the [format] script I've created (approximately [length] minutes):

[Display the script]

Would you like me to:
1. Generate the audio now
2. Make changes to the script
3. Adjust the length or tone

Step 5: Handle User Feedback

If user requests changes:

• Regenerate the script with adjustments
• Maintain the target word count
• Present the revised version

If user approves:

• Proceed to audio generation

Step 6: Generate Audio

Format the script for TTS:

1. Remove any remaining markdown (headers, bold, italics)
2. Ensure voice effects are in proper

   [effect]

   format
3. Check that pauses are appropriately placed
4. Verify numbers and acronyms are spelled out

Invoke the TTS script:

IMPORTANT: The

ELEVENLABS_API_KEY

environment variable is already configured in the system. Simply invoke the TTS script directly.

uv run /home/clawdbot/clawdbot/skills/sag/scripts/tts.py \
  -o /tmp/audio-gen-[timestamp]-[topic-slug].mp3 \
  -m eleven_multilingual_v2 \
  "[formatted_script]"

For long scripts, use heredoc:

uv run /home/clawdbot/clawdbot/skills/sag/scripts/tts.py \
  -o /tmp/audio-gen-[timestamp]-[topic-slug].mp3 \
  -m eleven_multilingual_v2 \
  "$(cat <<'EOF'
[formatted_script]
EOF
)"

Return the result:

MEDIA:/tmp/audio-gen-[timestamp]-[topic-slug].mp3

Your [format] is ready! [Brief description of content]. Duration: approximately [X] minutes.

Voice Effects (SSML Tags)

Available voice modulation effects (use sparingly for impact):

• [whispers]

  - Soft, intimate delivery

• [shouts]

  - Loud, emphatic delivery

• [excited]

  - Enthusiastic, energetic tone

• [serious]

  - Grave, solemn tone

• [sarcastic]

  - Ironic, mocking tone

• [sings]

  - Musical, melodic delivery

• [laughs]

  - Amused, jovial tone

• [short pause]

  - Brief silence (~0.5s)

• [long pause]

  - Extended silence (~1-2s)

Best Practices:

• Use effects for emotional moments, not every sentence
• Pauses are your most powerful tool for pacing
• Voice effects work best in audiobooks and dramatic content
• Keep podcasts and educational content mostly natural

Error Handling

Script Too Long

If the generated script exceeds target by >20%:

The script I generated is [X] words ([Y] minutes), which is longer than your target of [Z] minutes. Would you like me to:
1. Condense it to fit the target length
2. Split it into multiple parts
3. Keep it as is

Script Too Short

If the generated script is under target by >20%:

The script is [X] words ([Y] minutes), shorter than your target. Would you like me to:
1. Expand it with more detail
2. Add additional examples or stories
3. Generate as is

TTS Generation Fails

If the TTS script fails:

I've created the script, but I'm unable to generate the audio right now. Here's your script:

[Display script]

Error: [specific error message]

You can:
1. Check that ELEVENLABS_API_KEY is configured
2. Use the script with your own text-to-speech tool
3. Try again in a moment
4. Ask me to troubleshoot the audio generation

Common TTS Issues:

• API key not set: Verify ELEVENLABS_API_KEY in config
• Rate limit: Wait a moment and try again
• Text too long: Break into smaller chunks (max ~5000 characters)

Invalid Request

For unrealistic requests (e.g., "100-hour audiobook"):

That length would require [X] words and take significant time to generate. I recommend:
- Breaking it into multiple episodes/chapters
- Targeting 5-30 minutes per audio file
- Creating a series instead of one long file

Tips for Best Results

For Engaging Audiobooks

• Focus on character emotions and sensory details
• Use pauses to build dramatic tension
• Vary sentence length for rhythm
• Include internal monologue and reflection

For Compelling Podcasts

• Start with a question or surprising fact
• Use conversational phrases: "You know what's interesting..."
• Include relatable examples from everyday life
• End with actionable takeaways

For Effective Educational Content

• Use the "explain like I'm five" approach
• Build from simple to complex concepts
• Repeat key terms and definitions
• Provide multiple examples for clarity

Technical Notes

TTS Implementation:

• Uses Python script:

  ~/.clawdbot/clawdbot/skills/sag/scripts/tts.py

• No binary installation required (pure Python + requests)
• Directly calls ElevenLabs API
• Compatible with Linux and macOS

File Storage:

• Audio files are saved to

  /tmp/audio-gen/

• Filename format:

  audio-gen-[timestamp]-[topic-slug].mp3

• Files are automatically cleaned up after 24 hours

API Requirements:

• Anthropic API for script generation (already configured)
• ElevenLabs API for text-to-speech (configured via ELEVENLABS_API_KEY)
• Both services must be configured and have available credits

Supported Models:

• eleven_multilingual_v2

  - Best quality (default)

• eleven_turbo_v2

  - Faster generation

• eleven_turbo_v2_5

  - Fastest generation

• eleven_multilingual_v1

  - Legacy model

Cost Estimate:

• 10-minute audio (~750 words): approximately $1.43
  • Claude API: ~$0.075
  • ElevenLabs: ~$1.35
• Longer content scales proportionally

Generation Time:

• Script generation: 5-30 seconds (depending on length)
• Audio generation: 5-15 seconds (ElevenLabs processing)
• Total: Usually under 1 minute for 10-minute audio

Limitations

1. Maximum Length: 30 minutes (~2,250 words) per audio file

   • For longer content, create multiple parts/episodes

2. Single Voice: Currently supports one narrator voice

   • Cannot do multi-voice dialogue or character voices

3. No Background Music: Pure voice narration only

   • No background music, sound effects, or audio mixing

4. Real-time Generation: Each request generates fresh content

   • No pre-made templates or cached audio

5. Language: Primarily English

   • ElevenLabs supports other languages, but content generation optimized for English

Example Conversations

Example 1: Quick Audiobook

User: Create a 5-minute audiobook chapter about a lonely robot learning to feel emotions

Agent: I'll create a 5-minute sci-fi audiobook chapter. Let me write the story...

[Generates ~400-word narrative with emotional arc]

Here's the story I've created about Unit-7, a maintenance robot who discovers the meaning of friendship through an unexpected encounter. Would you like me to generate the audio or make changes first?

User: Generate the audio

Agent: [Invokes SAG with formatted script]

MEDIA:/tmp/audio-gen-1234567890-lonely-robot.mp3

Your audiobook chapter is ready! The story follows Unit-7's emotional awakening through a touching encounter with a young engineer. Duration: approximately 5 minutes.

Example 2: Podcast with Revisions

User: Make a 10-minute podcast about sustainable living tips

Agent: [Generates podcast script with intro, 5 practical tips, outro]

Here's a 10-minute podcast with 5 actionable sustainable living tips. The tone is encouraging and practical. Would you like me to:
1. Generate audio now
2. Make it more urgent/climate-focused
3. Add more specific examples
4. Adjust the length

User: Add more specific examples for each tip

Agent: [Regenerates with detailed examples]

Better? I've added specific examples like calculating your carbon footprint, composting at home, and choosing sustainable brands. Ready for audio?

User: Perfect!

Agent: MEDIA:/tmp/audio-gen-1234567890-sustainable-living.mp3

Your sustainable living podcast is ready with 5 practical tips, each with concrete examples you can start using today! Duration: 10 minutes.

Example 3: Educational Content

User: Create a 15-minute educational audio explaining quantum computing for beginners

Agent: I'll create a beginner-friendly 15-minute explanation of quantum computing...

[Generates ~1,100 words with analogies, examples, and clear structure]

Script preview:
- Introduction: What is quantum computing? (2 min)
- Classical vs Quantum: The key differences (4 min)
- Qubits and Superposition: How it works (5 min)
- Real Applications: What it means for you (3 min)
- Conclusion: The future of computing (1 min)

Should I generate the audio?

User: Yes please

Agent: MEDIA:/tmp/audio-gen-1234567890-quantum-computing.mp3

Your educational podcast on quantum computing is ready! It uses everyday analogies to explain qubits, superposition, and real-world applications. Duration: approximately 15 minutes.

When to Use This Skill

Use the audio-gen skill when users request:

• "Create an audiobook about..."
• "Make a podcast episode about..."
• "Generate audio content explaining..."
• "Record a story about..."
• "Produce an audio guide for..."
• "Create a narrated version of..."
• "Make an audio lesson about..."

Do NOT use for:

• Transcribing existing audio (use whisper skills instead)
• Converting user-written text to audio (use SAG skill directly)
• Real-time voice interaction (use voice-call plugin)
• Multi-speaker dialogues or interviews (current limitation)

Advanced Usage

Voice Selection

Users can request specific voices:

User: Create a podcast with the Rachel voice

Agent: [Checks voice configuration for "Rachel" voice ID and uses it in SAG command]

Series/Episodes

For multi-part content:

User: Create a 3-episode series about space exploration

Agent: I'll create Episode 1 first. Each episode will be 10 minutes. Let's start with "The Dawn of the Space Age"...

[After completion]

Episode 1 is ready! Would you like me to continue with Episode 2?

Format Blending

Mix formats for unique styles:

User: Create an educational podcast that tells a story

Agent: [Generates content that combines storytelling narrative with educational explanations]

Troubleshooting

Issue: Audio sounds robotic or unnatural Solution: Add more pauses and voice effects. Use contractions and conversational language.

Issue: Script doesn't match requested length Solution: Regenerate with explicit word count target. Check calculations (75 words/min).

Issue: Content is too technical or too simple Solution: Ask user for target audience. Adjust complexity accordingly.

Issue: SAG command fails Solution: Check ELEVENLABS_API_KEY is set. Verify SAG skill is installed and working.

Issue: User wants to edit the script manually Solution: Provide the plain text script. User can modify it and paste back for audio generation.

---

💡 Pro Tip: Always generate the script first and get user approval before creating audio. This saves time and API costs, and ensures the user gets exactly what they want.