tonitangpotato/agent-voice-skill/agent-voice-skill

Agent Voice

Send voice replies from your agent via Telegram (or any channel that supports audio).

Setup (one-time)

Run the setup script to install edge-tts:

bash scripts/setup.sh

This creates a Python venv at ~/.openclaw/tts-env and installs edge-tts. Also needs ffmpeg (brew/apt).

Generate voice

bash scripts/tts.sh "你好，这是语音测试" /tmp/voice.ogg
bash scripts/tts.sh "Hello world" /tmp/voice.ogg en-US-BrianMultilingualNeural

Arguments: tts.sh "text" [output_path] [voice_name]

Default voice: en-US-EmmaMultilingualNeural (neutral, clear, multilingual).

Send as Telegram voice message

Use the message tool:

{ "action": "send", "channel": "telegram", "to": "<chat_id>", "media": "/tmp/voice.ogg", "asVoice": true }

Key: asVoice: true sends as voice bubble, not file attachment.

Recommended voices

List all voices: ~/.openclaw/tts-env/bin/edge-tts --list-voices

Record preference in TOOLS.md

After choosing a voice, add to your TOOLS.md:

### TTS
-Voice: en-US-EmmaMultilingualNeural

Pipeline

Agent text → edge-tts (MP3) → ffmpeg (Opus/OGG) → Telegram voice note

Total latency: ~1-2s. Cost: $0 (edge-tts uses Microsoft's free TTS API).