Simple sound-to-text skill locally

Local speech-to-text using OpenAI Whisper. Use when the user needs to: (1) transcribe audio files to text, (2) convert voice messages to written content, (3)...

STT Simple - Local Speech-to-Text

🎯 触发场景 / Trigger Scenarios

使用此技能当用户需要 / Use this skill when user needs to:

1. 转录音频文件 / Transcribe audio files - 将 .ogg, .wav, .mp3 等格式转为文字 / Convert .ogg, .wav, .mp3 to text
2. 处理语音消息 / Process voice messages - WhatsApp/Telegram 语音消息转文字 / Voice message to text
3. 多语言识别 / Multi-language recognition - 支持中文、英文、日文等 99+ 语言 / Supports Chinese, English, Japanese, etc. (99+ languages)
4. 批量转录 / Batch transcription - 处理多个音频文件 / Process multiple audio files
5. 多 Agent 协作 / Multi-Agent collaboration - 多个 Agent 同时转录，输出隔离 / Multiple Agents transcribe simultaneously with isolated outputs

---

🚀 执行流程 / Execution Flow

1️⃣ 检查安装状态 / Check Installation

/root/.openclaw/venv/stt-simple/bin/whisper --version

如果未安装，先运行安装脚本 / If not installed, run install script first:

/root/.openclaw/workspace/skills/stt-simple/scripts/install.sh

2️⃣ 选择模型 / Model Selection

模型 / Model	大小 / Size	速度 / Speed	精度 / Accuracy	推荐场景 / Recommended For
tiny	39MB	⚡⚡⚡	⭐⭐⭐	快速测试 / Quick testing
base	74MB	⚡⚡	⭐⭐⭐⭐	日常使用 / Daily use
small	244MB	⚡	⭐⭐⭐⭐⭐	默认推荐 / Default
medium	769MB	🐌	⭐⭐⭐⭐⭐	高精度需求 / High accuracy
large	1.5GB	🐌🐌	⭐⭐⭐⭐⭐+	最佳质量 / Best quality

3️⃣ 执行转录 / Execute Transcription

方法 A: 使用 Whisper 命令行 / Use Whisper CLI

/root/.openclaw/venv/stt-simple/bin/whisper <audio_file> --model small --language Chinese

方法 B: 使用 Python 脚本（推荐支持多 Agent） / Use Python Script (Recommended for Multi-Agent)

# Without session isolation / 无会话隔离
/root/.openclaw/venv/stt-simple/bin/python \
  /root/.openclaw/workspace/skills/stt-simple/scripts/stt_simple.py \
  <audio_file> small zh

# With session isolation / 带会话隔离（多 Agent 场景）
/root/.openclaw/venv/stt-simple/bin/python \
  /root/.openclaw/workspace/skills/stt-simple/scripts/stt_simple.py \
  <audio_file> small zh agent-main-whatsapp

4️⃣ 输出位置 / Output Location

• 文本输出 / Text output: /root/.openclaw/workspace/stt_output/<session_id>/<filename>_<timestamp>.txt
• JSON 输出 / JSON output: 包含时间戳和置信度（如需）/ Includes timestamps and confidence (if needed)

---

🌍 语言代码 / Language Codes

语言 / Language	代码 / Code	别名 / Alias
中文 / Chinese	zh	Chinese
英文 / English	en	English
日文 / Japanese	ja	Japanese
韩文 / Korean	ko	Korean
法文 / French	fr	French
德文 / German	de	German
西班牙文 / Spanish	es	Spanish

自动检测 / Auto-detect: 省略 --language 参数 / Omit --language parameter

---

📁 输出格式 / Output Formats

• .txt - 纯文本 / Plain text (default)
• .json - 完整结果（含时间戳、置信度）/ Full results (with timestamps, confidence)
• .srt - 字幕格式（视频用）/ Subtitle format (for videos)
• .vtt - WebVTT（网页用）/ WebVTT (for web)

---

🔧 故障排查 / Troubleshooting

检查安装 / Check Installation

/root/.openclaw/venv/stt-simple/bin/whisper --version

重新安装 / Reinstall

rm -rf /root/.openclaw/venv/stt-simple
/root/.openclaw/workspace/skills/stt-simple/scripts/install.sh

手动下载模型 / Manual Model Download

/root/.openclaw/venv/stt-simple/bin/python \
  -c "import whisper; whisper.load_model('small')"

---

📦 资源文件 / Resources

文件 / File	路径 / Path	用途 / Purpose
安装脚本 / Install script	scripts/install.sh	一键安装虚拟环境、依赖、模型 / One-click install venv, dependencies, models
Python 脚本 / Python script	scripts/stt_simple.py	简化的转录 API，返回 JSON 结果 / Simplified transcription API with JSON output

---

💡 使用示例 / Examples

示例 1: 转录中文语音消息 / Transcribe Chinese voice message

/root/.openclaw/venv/stt-simple/bin/whisper \
  /root/.openclaw/media/inbound/voice.ogg \
  --model small --language Chinese

示例 2: 转录英文会议录音 / Transcribe English meeting recording

/root/.openclaw/venv/stt-simple/bin/whisper \
  meeting.wav --model medium --language en

示例 3: 使用 Python API 获取 JSON 结果 / Use Python API for JSON output

/root/.openclaw/venv/stt-simple/bin/python \
  /root/.openclaw/workspace/skills/stt-simple/scripts/stt_simple.py \
  audio.ogg small zh

示例 4: 多 Agent 场景 - 带会话隔离 / Multi-Agent with Session Isolation

# Jari (WhatsApp) - outputs to /root/.openclaw/workspace/stt_output/agent-jari-whatsapp/
/root/.openclaw/venv/stt-simple/bin/python \
  /root/.openclaw/workspace/skills/stt-simple/scripts/stt_simple.py \
  voice_a.ogg small zh agent-jari-whatsapp

# Other Agent (Telegram) - outputs to /root/.openclaw/workspace/stt_output/agent-telegram/
/root/.openclaw/venv/stt-simple/bin/python \
  /root/.openclaw/workspace/skills/stt-simple/scripts/stt_simple.py \
  voice_b.ogg small zh agent-telegram

---

🔐 多 Agent 支持 / Multi-Agent Support

为什么需要会话隔离 / Why Session Isolation?

当多个 Agent 同时使用 STT 功能时：

• 避免输出文件冲突 / Avoid output file conflicts
• 每个 Agent 可以追踪自己的转录结果 / Each Agent can track its own transcriptions
• 便于清理和管理 / Easier cleanup and management

会话 ID 命名建议 / Session ID Naming Suggestions

Agent / 场景	推荐 session_id	输出目录 / Output Directory
Jari (WhatsApp)	agent-jari-whatsapp	stt_output/agent-jari-whatsapp/
Eric (WhatsApp)	agent-eric-whatsapp	stt_output/agent-eric-whatsapp/
Telegram Agent	agent-telegram	stt_output/agent-telegram/
临时会话	session-<uuid>	stt_output/session-<uuid>/
用户专属	user-<user_id>	stt_output/user-<user_id>/

输出文件命名规则 / Output File Naming

<audio_filename>_<unique_timestamp>.txt

例如 / For example:

• voice_a_3f8b2c1d.txt
• meeting_9a4e7f2b.txt

每个文件名包含唯一的时间戳后缀，即使同一音频多次转录也不会覆盖。 Each filename includes a unique timestamp suffix, preventing overwrites even for repeated transcriptions.

---

⚠️ 注意事项 / Notes

1. CPU vs GPU: 默认使用 CPU，FP16 会自动降级为 FP32 / Defaults to CPU, FP16 auto-downgrades to FP32
2. 首次运行 / First run: 首次使用会下载模型（small 约 244MB）/ Downloads model on first use (~244MB for small)
3. 输出目录 / Output directory: 结果保存在 /root/.openclaw/workspace/stt_output/ / Results saved to /root/.openclaw/workspace/stt_output/
4. 隐私安全 / Privacy: 所有处理在本地完成，音频文件不会上传 / All processing is local, audio files never uploaded
5. 多 Agent / Multi-Agent: 建议使用 session_id 参数隔离输出 / Recommended to use session_id parameter for output isolation

---

🔑 默认配置 / Default Configuration

当前会话标识符 / Current Session ID:

agent-jari-whatsapp

输出目录 / Output Directory:

/root/.openclaw/workspace/stt_output/agent-jari-whatsapp/

快速调用 / Quick Start:

# 转录当前 WhatsApp 语音消息 / Transcribe current WhatsApp voice message
/root/.openclaw/venv/stt-simple/bin/python \
  /root/.openclaw/workspace/skills/stt-simple/scripts/stt_simple.py \
  <audio_file> small zh agent-jari-whatsapp