notebooklm-enterprise-api
编号. 通过Google Cloud使用NotebookLM Enterprise API生成播客音频。自动化手动NotebookLM工作流程 - 创建笔记本、上传源、使用自定义焦点提示生成音频。需要启用Discovery Engine API的Google Cloud项目。
📖 中文导读:编号. 通过Google Cloud使用NotebookLM Enterprise API生成播客音频。自动化手动NotebookLM工作流程 - 创建笔记本、上传源、使用自定义焦点提示生成音频。需要启用Discovery Engine API的Google Cloud项目。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。
name: notebooklm-enterprise-api description: Generate podcast audio using NotebookLM Enterprise API via Google Cloud. Automates the manual NotebookLM workflow - creates notebook, uploads sources, generates audio with custom focus prompt. Requires Google Cloud project with Discovery Engine API enabled.
NotebookLM Enterprise API Audio Generation
Generate podcast audio using the NotebookLM Enterprise API. This automates the manual NotebookLM web workflow:
- Create notebook via API
- Upload 5 source files (p1-brief.md, report.md, p3-briefing.md, sources.md, content_plan.md)
- Generate audio overview with the standard Yudame Research prompt
- Download the resulting MP3
Prerequisites
Google Cloud Setup
Enable APIs:
gcloud services enable discoveryengine.googleapis.com gcloud services enable aiplatform.googleapis.comRequired IAM Roles:
roles/discoveryengine.admin(for notebook creation)- Or appropriate NotebookLM Enterprise permissions
Authentication:
gcloud auth login gcloud config set project YOUR_PROJECT_IDNotebookLM 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
| Argument | Required | Description |
|---|---|---|
episode_dir |
Yes | Path to episode directory |
--series |
No | Series name (included in audio intro) |
--title |
No | Episode title (defaults to directory name) |
--cleanup |
No | Delete notebook after generation |
--timeout |
No | Timeout in minutes (default: 30) |
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
- Creates Notebook: API call to create a new notebook
- Uploads Sources: Uploads all 5 source files as text content
- Generates Audio: Calls
audioOverviews.createwith the standard Yudame Research prompt - Waits for Completion: Polls status every 30 seconds (up to timeout)
- Downloads MP3: Saves to
episode-directory/SLUG.mp3 - 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_IDfor 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 60for 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
| Operation | Endpoint |
|---|---|
| Create Notebook | POST /notebooks |
| Upload Source | POST /notebooks/{id}/sources:batchCreate |
| Generate Audio | POST /notebooks/{id}/audioOverviews |
| Check Status | GET /notebooks/{id}/audioOverviews/default |
| Download Audio | GET /notebooks/{id}/audioOverviews/default:download |
| Delete Notebook | DELETE /notebooks/{id} |