NotebookLM Enterprise API Audio Generation

Generate podcast audio using the NotebookLM Enterprise API. This automates the manual NotebookLM web workflow:

1. •Create notebook via API
2. •Upload 5 source files (p1-brief.md, report.md, p3-briefing.md, sources.md, content_plan.md)
3. •Generate audio overview with the standard Yudame Research prompt
4. •Download the resulting MP3

Prerequisites

Google Cloud Setup

1. •

   Enable APIs:

   bash

   gcloud services enable discoveryengine.googleapis.com
   gcloud services enable aiplatform.googleapis.com
   

2. •

   Required IAM Roles:

   • •roles/discoveryengine.admin (for notebook creation)
   • •Or appropriate NotebookLM Enterprise permissions
3. •

   Authentication:

   bash

   gcloud auth login
   gcloud config set project YOUR_PROJECT_ID
   

4. •

   NotebookLM Enterprise License:

   • •Must have Gemini Enterprise subscription with NotebookLM Enterprise enabled
   • •Access may be restricted to select customers

Environment Variables

# Optional - defaults to gcloud config
export GOOGLE_CLOUD_PROJECT=your-project-id

Usage

Command Line

cd podcast/tools

# Basic usage
python notebooklm_api.py ../episodes/YYYY-MM-DD-slug/

# With series name
python notebooklm_api.py ../episodes/cardiovascular-health/ep5-diet/ \
    --series "Cardiovascular Health"

# With custom title and cleanup
python notebooklm_api.py ../episodes/YYYY-MM-DD-slug/ \
    --title "Diet and Heart Health" \
    --series "Cardiovascular Health" \
    --cleanup

Arguments

Required Source Files

The episode directory must contain:

episode-directory/
├── research/
│   ├── p1-brief.md      # Phase 1 research brief
│   └── p3-briefing.md   # Master briefing (cross-validated)
├── report.md            # Narrative synthesis
├── sources.md           # Validated source links
└── content_plan.md      # Episode structure and NotebookLM guidance

What It Does

1. •Creates Notebook: API call to create a new notebook
2. •Uploads Sources: Uploads all 5 source files as text content
3. •Generates Audio: Calls audioOverviews.create with the standard Yudame Research prompt
4. •Waits for Completion: Polls status every 30 seconds (up to timeout)
5. •Downloads MP3: Saves to episode-directory/SLUG.mp3
6. •Cleanup: Optionally deletes the notebook

The Episode Focus Prompt

Uses the same prompt as the manual NotebookLM workflow:

• •Opening: "Yudame Research" + series name
• •Core principles: Spell out acronyms, define terms, cite studies
• •Tone: Intellectually rigorous but accessible
• •Closing: Summary + website URL

See generate_episode_focus() in the script for the full prompt.

Output

episode-directory/
└── YYYY-MM-DD-slug.mp3   # Generated audio (typically 20-40 min)

The audio is in the standard NotebookLM format (two hosts, Deep Dive style).

Troubleshooting

"Permission Denied"

• •Check gcloud authentication: gcloud auth print-access-token
• •Verify Discovery Engine API is enabled
• •Confirm NotebookLM Enterprise license is active

"Notebook Not Found"

• •The API uses project number, not project ID
• •Check gcloud projects describe PROJECT_ID for the number

"Audio Generation Failed"

• •Check source file sizes (must be under 100,000 tokens combined)
• •Verify source files are valid text/markdown
• •Check Google Cloud quotas

Timeout

• •Increase with --timeout 60 for larger source files
• •Audio generation typically takes 5-15 minutes

Integration with Workflow

This is the primary audio generation method for Phase 10.

Fallback: Manual NotebookLM web interface (.claude/skills/notebooklm-audio/) when API is unavailable.

API Reference

Endpoints Used

Documentation

• •Audio Overview API
• •Notebook API
• •Source Management