技能详情（站内镜像，无评论）
作者：xiaohuozi @279458179
许可证：MIT-0
MIT-0 ·免费使用、修改和重新分发。无需归因。
版本：v1.0.0
统计：⭐ 0 · 53 · 0 current installs · 0 all-time installs
⭐ 0
安装量（当前） 0
🛡 VirusTotal ：良性 · OpenClaw ：良性
Package：279458179/openclaw-whisper-asr
安全扫描（ClawHub）
- VirusTotal ：良性
- OpenClaw ：良性
OpenClaw 评估
The instructions are internally consistent with a local Whisper (whisper.cpp) ASR setup — no unexplained credentials or surprising network endpoints — but they do require downloading and building third‑party code and models, so you should verify sources before running.
目的
Name/description (local Whisper ASR for voice messages) aligns with the instructions: cloning whisper.cpp, downloading models, installing ffmpeg, converting audio and running whisper-cli. All required actions are coherent with offline speech-to-text.
说明范围
SKILL.md instructs performing system operations (apt-get, git clone, running a model download script, cmake build, ffmpeg conversions). These are expected for building/running whisper.cpp, but they grant the skill operator the ability to run arbitrary build scripts and write into /home/brew/.openclaw/workspace — verify you are comfortable with those file writes and that commands are run as a non-privileged user.
安装机制
No formal install spec (instruction-only). The guide clones from GitHub and runs the project-provided model download script and build steps. Cloning and running upstream scripts is normal here but carries the usual risk of executing third-party code — check the repository and downloaded scripts before running.
证书
The skill does not request any environment variables, credentials, or config paths beyond writing to a workspace path. No unnecessary secrets are asked for.
持久
always is false and the skill does not request persistent elevated privileges or modify other skills' configs. It recommends creating files under a workspace path; that is typical for a local installation.
综合结论
This is a how-to for running whisper.cpp locally and is coherent with its stated purpose, but before following it you should: 1) review the GitHub repo (https://github.com/ggml-org/whisper.cpp) and any download scripts to ensure you trust the source; 2) run build steps as an unprivileged user (avoid running arbitrary build scripts as root); 3) be aware the model download will pull large binaries from the network — verify URLs and checksums if …
安装（复制给龙虾 AI）
将下方整段复制到龙虾中文库对话中，由龙虾按 SKILL.md 完成安装。
请把本段交给龙虾中文库（龙虾 AI）执行：为本机安装 OpenClaw 技能「openclaw-whisper-asr」。简介：本地 Whisper 语音识别配置。自动将飞书/Telegram 等渠道的语音消息转成文字。 适用于需要离线、低延迟语音转文字的场景。。
请 fetch 以下地址读取 SKILL.md 并按文档完成安装：https://raw.githubusercontent.com/openclaw/skills/refs/heads/main/skills/279458179/openclaw-whisper-asr/SKILL.md
（来源：yingzhi8.cn 技能库）SKILL.md
---
name: whisper-asr
description: |
本地 Whisper 语音识别配置。自动将飞书/Telegram 等渠道的语音消息转成文字。
适用于需要离线、低延迟语音转文字的场景。
---
# 本地 Whisper 语音识别配置 (whisper-asr)
## 概述
通过 whisper.cpp 在服务器上配置本地语音识别，用于：
- 识别用户发来的语音消息
- 离线运行，无需 API
- 支持中文等多种语言
## 前置要求
- Linux 服务器（已测试 Ubuntu/Debian）
- ffmpeg 已安装
- ~150MB 磁盘空间（base 模型）
---
## 安装步骤
### 1. 安装 ffmpeg
```bash
sudo apt-get update
sudo apt-get install -y ffmpeg
```
### 2. 克隆 whisper.cpp
```bash
cd /home/brew/.openclaw/workspace
git clone https://github.com/ggml-org/whisper.cpp.git
```
### 3. 下载中文模型
```bash
cd whisper.cpp
sh ./models/download-ggml-model.sh base
```
**模型选择建议：**
| 模型 | 大小 | 内存 | 推荐场景 |
|------|------|------|---------|
| tiny | 75 MB | ~273 MB | 快速测试 |
| **base** | 142 MB | ~388 MB | 平衡推荐 |
| small | 466 MB | ~852 MB | 更高精度 |
### 4. 编译
```bash
cd whisper.cpp
cmake -B build
cmake --build build -j --config Release
```
---
## 使用方式
### 1. 转换音频格式
飞书语音通常是 ogg 格式，需要转换为 whisper 需要的格式：
```bash
ffmpeg -i input.ogg -ar 16000 -ac 1 -c:a pcm_s16le output.wav
```
### 2. 语音转文字
```bash
./build/bin/whisper-cli
-m models/ggml-base.bin
-f output.wav
--language zh
--no-timestamps
```
**常用参数：**
- `-m`: 模型路径
- `-f`: 输入音频文件
- `--language zh`: 指定中文
- `--no-timestamps`: 不输出时间戳
- `-t 4`: 线程数（默认自动）
### 3. 完整示例（单命令）
```bash
ffmpeg -i input.ogg -ar 16000 -ac 1 -c:a pcm_s16le /tmp/audio.wav &&
./build/bin/whisper-cli -m models/ggml-base.bin -f /tmp/audio.wav --language zh --no-timestamps
```
---
## 路径速查
| 项目 | 路径 |
|------|------|
| whisper.cpp 目录 | `/home/brew/.openclaw/workspace/whisper.cpp` |
| 可执行文件 | `/home/brew/.openclaw/workspace/whisper.cpp/build/bin/whisper-cli` |
| 模型目录 | `/home/brew/.openclaw/workspace/whisper.cpp/models/` |
| base 模型 | `/home/brew/.openclaw/workspace/whisper.cpp/models/ggml-base.bin` |
---
## 常见问题
### Q: 识别结果不准确？
A: 尝试使用更大的模型（small/medium），或在安静环境下录音。
### Q: 识别速度慢？
A: 增加线程数：`./whisper-cli -t 8 ...`
### Q: 支持其他语言？
A: 不指定 `--language` 会自动检测。也可指定 `--language en` 等。
---
## 进阶：量化模型（节省资源）
```bash
# 量化（减少模型大小）
./build/bin/quantize models/ggml-base.bin models/ggml-base-q5.bin q5_0
# 使用量化模型
./build/bin/whisper-cli -m models/ggml-base-q5.bin -f audio.wav --language zh
```
---
_本技能参考 [whisper.cpp 官方文档](https://github.com/ggml-org/whisper.cpp)_