openclaw-speaches-stt

Speaches realtime speech-to-text provider for OpenClaw voice-call streaming.

This plugin registers a speaches realtime transcription provider that connects to Speaches' OpenAI-compatible realtime endpoint and can convert Twilio μ-law 8k audio to PCM16 16k before forwarding audio.

Example OpenClaw voice-call config

{
  "streaming": {
    "enabled": true,
    "provider": "speaches",
    "providers": {
      "speaches": {
        "baseUrl": "http://127.0.0.1:8000/v1",
        "model": "Systran/faster-distil-whisper-small.en",
        "apiKey": "cant-be-empty",
        "silenceDurationMs": 500,
        "vadThreshold": 0.5,
        "prefixPaddingMs": 0,
        "convertTwilioMulaw": true
      }
    }
  }
}

Environment variables

• SPEACHES_BASE_URL — default http://127.0.0.1:8000/v1
• SPEACHES_STT_MODEL — default Systran/faster-distil-whisper-small.en
• SPEACHES_API_KEY — default cant-be-empty

Notes

For latency-sensitive phone calls on CPU, small/medium faster-whisper models are usually a better fit than large models.