ai-podcast-production

# AI Podcast Production Capability Pack > Cross-agent portable judgment for AI-produced podcasts. Covers the full pipeline: article analysis to script writing with adversarial Codex review, large-chunk TTS generation with inline denoising, dual-BGM selection and envelope-follower arrangement, show notes production, and Colab deployment patterns. Grounded in 2 episodes produced, 10+ arrangement iterations, 25+ Codex corrections validated. **CONSUMES**: Source articles/books (text), reference voice audio (WAV), LoRA weights (safetensors), BGM tracks (WAV/MP3). **PRODUCES**: Podcast script (text), episode audio (WAV 48kHz), show notes (markdown), Colab notebooks (.ipynb). > **INTERFACE**: ai-voice-production pack defers to this pack for podcast-specific TTS chunking strategy, arrangement algorithms, and script methodology. This pack defers to ai-voice-production for TTS tool selection and voice cloning method choice. ml-training pack provides platform selection and cost estimation for cloud GPU. This pack defers cloud platform details to ml-training's platform-selection.md. When all three packs load: ml-training takes precedence for platform/cost; ai-voice-production takes precedence for tool selection; this pack takes precedence for podcast-specific workflow (chunking, arrangement, script quality, show notes). --- ## Step 0: Pack Prerequisites - **Google Colab Pro** (or free tier with gotcha awareness) — TTS requires A100 GPU - **Python 3.10+** with: `voxcpm`, `soundfile`...