Review & Update: Podcast Audio Engine PRD

[valorengels]

📄 Plan: docs/plans/podcast-audio.md

---

Summary

This PRD document (docs/plans/podcast-audio.md) outlines technical and creative requirements for generating podcast audio using Gemini TTS API with full script control.

Key topics covered:

• Target: 35-minute episodes at ~5,200 words
• Text-first approach: Script with embedded TTS directives -> Gemini TTS
• Model configuration: gemini-2.5-flash-preview-tts with Alnilam voice
• Script structure with directive syntax (VOICE, PACE, PAUSE, EMPHASIS, TRANSITION)
• Voice identity mapping from VOICE-IDENTITY.md
• Audio generation pipeline with section splitting and stitching
• Implementation code for generate_audio_tts.py
• Quality assurance checklists
• Cost estimation (~$0.10-0.20 per episode)

Review Needed

This plan needs review and potential updates after recent episode creator improvements (Waves 1-5 implemented).

Key sections that may need revision:

1. Architecture Overview (lines 23-83) - Verify pipeline matches current implementation
2. TTS API Configuration (lines 89-138) - Check if model IDs and voices are still current
3. Script Structure Template (lines 250-412) - Compare with current content_plan.md template
4. Audio Generation Pipeline (lines 416-444) - Verify Phase 9 entry requirements match current workflow
5. Implementation code (lines 459-613) - Check if generate_audio_tts.py was created and is current
6. File Structure (lines 664-679) - Verify alignment with current episode directory structure
7. Environment Setup (lines 683-710) - Confirm dependencies and API key handling

Questions to address:

• Is Gemini TTS the actual audio generation method being used, or is NotebookLM still primary?
• Has generate_audio_tts.py been implemented and tested?
• Do the directive categories match what Gemini TTS actually supports?
• How does this integrate with the Wave 1-5 improvements to content planning?
• What is the relationship between script.md and content_plan.md?