Using speech-recognition Skill to Make OpenClaw Understand What You Say

A step-by-step guide to configuring the speech-recognition skill in OpenClaw, using SiliconFlow's SenseVoice API for automatic transcription of multi-format voice messages, with direct Telegram voice message support.

Why speech-recognition?

The most natural way to communicate is speaking. The speech-recognition skill bridges your voice and OpenClaw's text processing — letting you send voice messages in Telegram (or other channels) and have them automatically transcribed and processed by your Agent.

What It Supports

• Telegram voice messages (direct processing)
• Audio file uploads (MP3, WAV, OGG, M4A)
• Multi-language recognition (Chinese, English, Japanese, Korean, and more)
• Automatic language detection

Setup: SiliconFlow SenseVoice API

clawhub install speech-recognition

SenseVoice from SiliconFlow is recommended — good Chinese recognition, generous free tier, easy API setup. Register at siliconflow.cn and get your API key.

Configure in OpenClaw:

{
  "speech": {
    "provider": "siliconflow",
    "apiKey": "your-siliconflow-key",
    "model": "SenseVoiceSmall"
  }
}

How It Works in Practice

Once configured, send a voice message to OpenClaw in Telegram. The skill automatically:

1. Detects the voice message type
2. Downloads the audio file
3. Sends to SenseVoice API for transcription
4. Returns transcribed text to the Agent
5. Agent processes the text request normally

The whole flow is invisible to you — you just speak, OpenClaw responds as if you typed it.

Real-World Uses at SFD Lab

On-the-go task assignment: Walking between meetings, voice-assign tasks to the dispatcher Agent. No need to stop and type.

Quick content drafts: Dictate a rough idea for an article — the transcription goes to fox for structuring.

Voice-to-note capture: Combine with the writing skill to create voice-dictated notes that get formatted automatically.

Pitfalls

Background noise: SenseVoice handles moderate noise well, but loud environments reduce accuracy. Speak clearly for critical inputs.

Technical terms and proper nouns: AI-specific terms (like "OpenClaw" or "ClawHub") may be transcribed incorrectly. Add a custom vocabulary list if needed.

Audio format compatibility: Telegram sends voice as OGG — the skill handles conversion automatically. For other platforms, check supported formats.

API rate limits: SiliconFlow's free tier has limits. For heavy voice usage, monitor your quota.

Summary

If you use OpenClaw on mobile frequently, speech-recognition is a quality-of-life upgrade. Install it once and voice input just works.

---

SFD Lab | SenseVoice + OpenClaw voice input setup guide