If you're building AI agents (like Claude Code) that need to generate speech, the SKILL.md file provides agent-optimized documentation.
Copy the skill file to your agent's skills directory:
# For Claude Code
mkdir -p ~/.claude/skills/speak-tts
cp SKILL.md ~/.claude/skills/speak-tts/SKILL.md
# For Pi
mkdir -p ~/.pi/skills/speak-tts
cp SKILL.md ~/.pi/skills/speak-tts/SKILL.mdThe SKILL.md includes:
- Quick start commands
- Common workflow patterns (streaming, batch, long documents)
- All CLI options in a reference table
- Performance expectations
- Long document handling (auto-chunk, resume)
- Emotion tags reference
Long Document Reliability:
# Auto-chunk prevents timeouts on long content
speak book-chapter.md --auto-chunk --output chapter.wav
# Resume if interrupted
speak --resume manifest.json
# Estimate duration before committing
speak --estimate document.mdBatch Processing:
# Process multiple files
speak *.md --output-dir ~/Audio/ --skip-existingConcatenation:
# Combine audio files
speak concat part1.wav part2.wav --out combined.wav| Parameter | Default | Notes |
|---|---|---|
| Model | chatterbox-turbo-8bit |
Good balance of speed/quality |
| Temperature | 0.5 |
Balanced expressiveness |
| Speed | 1.0 |
Natural playback |
| Timeout | 300 |
5 minutes, use --auto-chunk for longer |
| Content | Recommendation |
|---|---|
| Short text (<500 chars) | speak "text" --play |
| Medium text | speak file.md --stream |
| Long documents (>10 min audio) | speak file.md --auto-chunk --output out.wav |
| Multiple files | speak *.md --output-dir dir/ |
See SKILL.md for complete documentation.