Meeting Transcriber

基于FunASR实现的中文实时会议语音转录，支持自动标点、转录保存、环境检测和文件管理。

Security Scan

The skill mostly does what its description says (a local Windows/Conda-based meeting transcriber) but contains several inconsistencies and assumptions — notably hard-coded user paths and a dependency on external project files that are not included — that make its behavior surprising and worth manual review before installing.

Purpose & Capability

The skill claims to provide an OpenClaw meeting transcriber, which matches the code, but the package is not self-contained: meta.json and the scripts expect a specific Windows/Conda setup and external files at D:\dev\python\voiceFunAsr (vocie_mic_fixed_gbk.py, meeting_records, etc.). The manifest given to you earlier stated no required binaries/env, but _meta.json lists required binaries ('python','conda') and external file paths. Requiring and invoking files outside the skill bundle (hard-coded D:\ and C:\Users\pengjschina paths) is disproportionate to a self-contained skill and is inconsistent with the declared requirements.

Instruction Scope

SKILL.md and the shipped scripts instruct the agent/user to read and create files in specific system locations (D:\dev\python\voiceFunAsr, C:\Users\pengjschina\.cache, and the user's home .openclaw directory), run conda commands, and possibly download models automatically. The instructions direct editing/writing of SKILL.md and create batch scripts; they rely on or manipulate files that are outside the skill's own files. This is broader than simply 'transcribe audio' and grants the skill filesystem and execution scope beyond a minimal transcriber.

Install Mechanism

There is no declared install spec (instruction-only), which reduces upfront install risk, but the included scripts (conda_setup.py, transcriber.py, openclaw_wrapper.py) will run subprocesses that can install packages, create files, and execute external scripts when invoked. No remote arbitrary archive downloads are present in the package, but runtime behavior may cause network model downloads (modelscope) and conda/pip installs.

Credentials

The skill requests no explicit environment variables or credentials, but it hard-codes and expects access to specific system locations (D:\ dev paths and a specific user cache at C:\Users\pengjschina). It also requires Conda and a particular conda env ('audioProject'). Requiring access to and modifying files in those user/system paths (including writing a config to ~/.openclaw) is not proportionate unless the user actually intends to integrate with that existing project layout.

Persistence & Privilege

always:false (good). The skill will write files under its directory and create a config file under the user's home (~/.openclaw), plus batch scripts (start_with_conda.bat / start_meeting.bat). The conda_setup.py includes logic to update SKILL.md and possibly insert code into a 'meeting_minutes.py' — modifying files in the skill directory is normal, but modifying other code files (if present) or writing to the user's config should be noted and approved by the user.

What to consider before installing

This skill is plausible for local Windows/Conda use, but there are red flags you should check before using it: - Inconsistency: the top-level metadata stated no required binaries, but _meta.json and the scripts expect 'python' and 'conda' and a Conda env named 'audioProject'. Decide whether you run this on Windows with Conda. - External dependencies: the core transcriber logic it tries to run (vocie_mic_fixed_gbk.py and other files) is expected at D:\dev\python\voiceFunAsr but those files are not included. Inspect or provide that project directory yourself; the skill will fail or may run unexpected code if those files differ. - Hard-coded paths: the scripts reference C:\Users\pengjschina and D:\ paths. If you run this on your machine those paths will differ; the scripts may create or modify files in your home directory (~/.openclaw) and in the working_directory. Back up anything important first. - Runtime actions: running the provided setup scripts will call conda/pip, create batch scripts, and may download ASR models from the network. Review conda_setup.py and the external transcriber scripts (vocie_mic_*) before running them, ideally in an isolated environment (VM or disposable account). - If you want to proceed: ensure you understand and control the external project files it depends on, run the environment checks manually, and consider running the code in a contained Windows/Conda environment to limit unintended filesystem/network effects. If you are unsure, do not run the setup/start commands until you verify the referenced project files and model download sources.

Meeting Transcriber - 会议转录技能

基于 FunASR 的实时会议转录技能，支持中文语音识别、实时转录、文件保存和环境管理。

🎯 功能特性

🎙️ 实时语音转录

• 实时麦克风录音和语音识别
• 支持中文语音识别
• 智能语音活动检测（VAD）
• 自动标点恢复

📝 会议纪要生成

• 自动保存转录文本
• 按时间戳记录发言
• 生成结构化会议纪要
• 支持导出文本格式

🔧 环境管理

• Conda 环境自动检测
• 依赖包完整性检查
• 一键环境配置
• 音频设备检测

🚀 快速开始

安装技能

技能已自动注册到 OpenClaw，无需额外安装。

使用命令

在 OpenClaw 会话中，使用以下命令：

转录操作

• 开始会议转录 - 启动实时语音转录
• 停止转录 - 停止转录并保存文件

环境管理

• 检查转录环境 - 检查 Conda 环境和依赖包
• 设置转录环境 - 配置 Conda 环境和安装依赖

文件管理

• 查看会议记录 - 列出所有会议转录文件
• 打开最新会议纪要 - 打开最新的会议纪要文件

📁 文件结构

meeting-transcriber/
├── _meta.json          # 技能元数据
├── transcriber.py      # 技能主文件（支持所有命令）
├── start_meeting.bat   # 批处理启动文件
├── check_env.py        # 环境检查工具
├── conda_setup.py      # 环境设置工具
└── SKILL.md           # 技能说明文档

🔧 技术要求

必需条件

• 操作系统: Windows 10/11
• Conda: Miniconda 或 Anaconda
• Python: 3.8+（通过 Conda audioProject 环境）
• Conda 环境: audioProject

依赖包

• funasr - 语音识别引擎
• torch - 深度学习框架
• pyaudio - 音频输入输出
• webrtcvad - 语音活动检测
• numpy - 数值计算

📊 使用示例

示例 1: 完整会议转录流程

用户: 开始会议转录
AI: 启动转录系统...
用户: （开始会议讨论）
用户: 停止转录
AI: 转录已保存，文件位置: ...

示例 2: 环境检查

用户: 检查转录环境
AI: 检查Python环境... OK
AI: 检查项目目录... OK
AI: 检查依赖包... OK
AI: 环境检查通过！

示例 3: 查看会议记录

用户: 查看会议记录
AI: 找到 5 个转录文件:
AI: 1. meeting_20260402_103149.txt
AI: 2. meeting_20260402_100724.txt
AI: ...

🛠️ 故障排除

常见问题

Q: 环境检查失败

A: 1. 运行'设置转录环境'命令
   2. 或手动创建环境: conda create -n audioProject python=3.8
   3. 手动安装依赖: pip install funasr torch pyaudio webrtcvad

Q: 转录启动失败

A: 1. 检查麦克风权限
   2. 运行'检查转录环境'验证环境
   3. 尝试使用批处理文件: start_meeting.bat

Q: 编码错误（乱码）

A: 1. 使用批处理文件: start_meeting.bat
   2. 脚本已内置编码修复

Q: 麦克风无法工作

A: 1. 检查麦克风是否被其他程序占用
   2. 运行环境检查查看音频设备
   3. 确保麦克风已正确连接

性能优化建议

1. 硬件优化

   • 使用 USB 麦克风或耳机麦克风
   • 确保足够的系统内存（建议 8GB+）
   • 使用 SSD 存储提高 IO 性能

2. 软件优化

   • 关闭不必要的后台程序
   • 调整音频缓冲区大小
   • 使用安静的录音环境

3. 环境优化

   • 在安静的房间中使用
   • 避免风扇、空调等背景噪音
   • 说话时靠近麦克风

📈 高级功能

自定义配置

技能支持以下自定义配置：

• 转录语言设置（默认中文）
• VAD 参数调整
• 保存路径配置
• 音频设备选择

扩展功能

如需扩展功能，可以修改以下文件：

• D:/dev/python/voiceFunAsr/vocie_mic_fixed_gbk.py - 核心转录逻辑
• transcriber.py - 技能命令处理逻辑

📝 文件说明

核心文件

文件	功能	使用场景
transcriber.py	技能主文件	处理所有命令，OpenClaw自动调用
start_meeting.bat	批处理启动文件	直接启动转录，解决编码问题
check_env.py	环境检查工具	独立运行环境检查
conda_setup.py	环境设置工具	自动安装依赖和环境

转录文件

• 位置: D:\dev\python\voiceFunAsr\meeting_records\
• 格式: meeting_YYYYMMDD_HHMMSS.txt
• 内容: 按时间戳记录的转录文本

🔄 更新日志

v1.0.0 (2026-04-02)

• 初始版本发布
• 基于 FunASR 的实时语音转录
• 完整的命令支持
• 环境检查和管理
• 文件查看和打开功能
• OpenClaw 技能集成

计划功能

• 多说话人识别
• 实时翻译功能
• 云端同步备份
• 移动端支持

📞 技术支持

如有问题，请提供：

1. 操作系统和 Python 版本
2. 错误日志或截图
3. 麦克风设备信息
4. 会议环境描述

---

提示: 首次使用时建议运行环境检查：

检查转录环境

转录文件位置: D:\dev\python\voiceFunAsr\meeting_records\