voice-setup

name: voice-setup description: Set up free voice functionality (TTS + STT) for OpenClaw using Edge TTS and whisper-cpp. Use when: (1) User wants to add voice/audio capabilities, (2) Setting up speech-to-text transcription, (3) Configuring text-to-speech synthesis, (4) Enabling voice messages on Telegram/WhatsApp, (5) User asks about free TTS/STT solutions without API keys.

Voice Setup Skill

This skill helps configure free, open-source voice capabilities for OpenClaw.

Overview

Prerequisites

Check and install dependencies:

# Check for Homebrew
which brew || echo "Install Homebrew first: https://brew.sh"

# Install whisper-cpp (STT)
brew install whisper-cpp

# Install ffmpeg (audio conversion)
brew install ffmpeg

# Download whisper model (base is recommended for balance of speed/accuracy)
mkdir -p ~/.openclaw/models
curl -L -o ~/.openclaw/models/ggml-base.bin \
  "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin"

Configuration

Apply this config patch to enable voice:

{
  "messages": {
    "tts": {
      "auto": "inbound",
      "provider": "edge",
      "edge": {
        "enabled": true,
        "voice": "zh-CN-XiaoxiaoNeural",
        "lang": "zh-CN",
        "outputFormat": "audio-24khz-48kbitrate-mono-mp3"
      }
    }
  },
  "tools": {
    "media": {
      "audio": {
        "enabled": true,
        "maxBytes": 20971520,
        "models": [
          {
            "type": "cli",
            "command": "bash",
            "args": ["-c", "ffmpeg -i '{{MediaPath}}' -ar 16000 -ac 1 /tmp/whisper-input.wav -y >/dev/null 2>&1 && whisper-cli -m ~/.openclaw/models/ggml-base.bin -l auto -f /tmp/whisper-input.wav 2>/dev/null | grep -E '^\\[' | sed 's/\\[.*\\] //'"],
            "timeoutSeconds": 120
          }
        ]
      }
    }
  }
}

TTS Voice Options

Chinese Voices

• zh-CN-XiaoxiaoNeural - 女声，活泼 (recommended)
• zh-CN-YunxiNeural - 男声，自然
• zh-CN-XiaoyiNeural - 女声，温柔

English Voices

• en-US-JennyNeural - Female, warm
• en-US-GuyNeural - Male, natural
• en-GB-SoniaNeural - British female

Other Languages

• ja-JP-NanamiNeural - Japanese female
• ko-KR-SunHiNeural - Korean female
• de-DE-KatjaNeural - German female

List all voices: npx node-edge-tts --list-voices

TTS Modes

Whisper Models

Testing

Test TTS

npx node-edge-tts -t '你好！语音测试' -v 'zh-CN-XiaoxiaoNeural' -f /tmp/test.mp3

Test STT

ffmpeg -i input.ogg -ar 16000 -ac 1 /tmp/test.wav -y
whisper-cli -m ~/.openclaw/models/ggml-base.bin -l auto -f /tmp/test.wav

Troubleshooting

"whisper-cli not found"

brew install whisper-cpp

"ffmpeg not found"

brew install ffmpeg

Audio file format error

Telegram sends OGG/Opus, whisper needs WAV. The config handles conversion automatically.

Model not found

curl -L -o ~/.openclaw/models/ggml-base.bin \
  "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin"