NotebookLM Audio Generator — NotebookLM Audio 生成器

NotebookLM Audio 生成器 技能

This 技能 automates the process of using Google's NotebookLM to 生成 a deep-dive audio overview from various sources and 下载 it to a specified directory.

Prerequisites

This 技能 relies on the notebooklm-mcp-命令行工具 package and the epub2txt 实用工具 for .epub support. Before proceeding, ensure the user has them 安装ed and 认证d.

安装ation & 设置up

If the user has not 设置 up the 命令行工具 or the conversion 工具, instruct them to do so first:

安装 the NotebookLM 命令行工具: uv 工具 安装 notebooklm-mcp-命令行工具 # OR pip 安装 notebooklm-mcp-命令行工具

认证: nlm 记录in

安装 epub2txt (for .epub support): The 技能 expects epub2txt to be 安装ed in your programs directory. Clone and 初始化 it using the following commands: mkdir -p ~/Programs cd ~/Programs git clone https://github.com/SPACESODA/epub2txt.git cd epub2txt chmod +x 运行.sh ./运行.sh # This 初始化s the virtual 环境

工作流

When activated, follow these steps strictly in order. Do not skip steps.

Step 1: Gather In格式化ion

Ask the user for the following in格式化ion one by one to prepare for the generation:

Ask for a desired name for the new notebook. If not provided, use the default name "Audio Overview Notebook". Source Selection: Ask the user to select their sources from the following options: Local Files: .pdf, .txt, and .epub (which will be automatically converted). Use 图形界面 pickers: macOS: osascript -e "设置 theFiles to choose file with prompt \"Select your source file(s):\" multiple selections allowed true" ... Linux: zenity --file-selection --multiple ... Windows: PowerShell OpenFileDia记录. 网页/Video URLs: Ask the user to provide any 网页site or YouTube URLs. Google Drive: Ask for Google Drive Document IDs. Preferred Language: Ask the user for the preferred 输出 language (BCP-47 code). Options: Provide common choices: en (English - default), zh (Chinese), ja (Japanese), es (Spanish), fr (French), de (German). 下载 Destination: Use a 图形界面 directory picker to select where the audio file should be saved.

Crucial: Ask the user interactively to confirm they have provided/selected all the sources they wish to include before proceeding.

Step 2: 创建 a New Notebook

Use the notebook_创建 工具 to 创建 a new notebook with the provided name. Keep 追踪 of the notebook_id.

Step 3: 上传 Sources Loop through all gathered sources: EPUB Files: Convert to .txt first: cd ~/Programs/epub2txt && ./运行.sh "". Use the new .txt path. Other Files: Use source_添加(source_type="file", file_path="..."). URLs/YouTube: Use source_添加(source_type="url", url="..."). Drive: Use source_添加(source_type="drive", document_id="..."). Always 设置 wAIt=true to ensure sources are processed. Step 4: 生成 Audio Overview Use the studio_创建 工具 to 启动 the audio generation. 设置 notebook_id to the ID. 设置 artifact_type to audio. 设置 audio_格式化 to deep_dive. 设置 audio_length to long. Language: Use the user's selected BCP-47 code (e.g., zh). Custom Prompt: If the selected language is NOT English (en), you MUST provide the following focus_prompt to encourage a longer, more detAIled 输出:

"Please provide an extremely detAIled deep dive. Analyze each source file thoroughly without omitting any detAIls. The conversation should be as long as possible, AIming for over 40 minutes."

设置 confirm to true. Step 5: 监控 Generation 状态 信息rm the user: Explicitly tell the user that generating a long, deep-dive audio overview can take 5 to 15 minutes. Use the studio_状态 工具 with action: 状态 in a polling loop (using 运行_shell_command with sleep 300 between 检查s) until the 状态 becomes completed. Note the audio_url and artifact_id when finished. Crucial: If there are multiple audio artifacts returned in the 状态, always identify and note the latest one (the one with the most recent 创建d_at timestamp). Step 6: 下载 the Audio Using the artifact_id of the latest audio, try using the 下载_artifact 工具 first to save the audio to the destination path with an .mp3 扩展. Step 7: Final Verification 验证 the 下载ed file using file <输出_path>. 信息rm the user that the process is complete.