Add text-to-speech audio generation for blog posts

[cduruk]

Summary

• Adds ability to generate spoken audio versions of blog posts using Modal + Chatterbox TTS
• TypeScript orchestrator handles file discovery, MDX text extraction, and Modal invocation
• Python Modal function runs Chatterbox TTS on GPU (A10G) for high-quality speech synthesis

Changes

• scripts/generate-audio.ts - TypeScript orchestrator with CLI
• scripts/tts/chatterbox_tts.py - Modal function with Chatterbox TTS
• package.json - Added generate-audio npm script
• AGENTS.md - Documentation for audio generation
• .gitignore - Added Python cache files

Usage

npm run generate-audio                    # Generate audio for posts without audio.wav
npm run generate-audio -- --slug my-post  # Generate for specific post(s)
npm run generate-audio -- --force         # Regenerate all posts

Technical Notes

• Text is cleaned (removes code blocks, components, special chars) and truncated to ~3000 chars
• CUDA errors occur with longer text - limit determined through testing
• Cold boot takes ~30 seconds on Modal, then fast generation
• Output is WAV format (MP3 conversion planned for future)

Requirements

• uv (Python package manager)
• Modal account with GPU access

Test plan

• Test TTS locally with short text
• Test end-to-end with blog post
• Verify WAV file is valid audio

🤖 Generated with Claude Code