Voice Assistant

Windows voice companion for OpenClaw. Custom wake word via Porcupine, local STT via faster-whisper, streamed responses over the gateway WebSocket, and ElevenLabs TTS with natural chime/thinking sounds. Supports multi-turn conversation with automatic follow-up listening, mic suppression to prevent feedback, and a system tray with pause/resume. Recommended voices: Matilda (XrExE9yKIg1WjnnlVkGX, free tier) or Ivy (MClEFoImJXBTgLwdLI5n, paid tier). Fully customizable wake word, voice, hotkey, and silence thresholds.

This skill appears to do what it says: it listens on your microphone, transcribes speech locally, sends text to your configured OpenClaw gateway, and uses ElevenLabs to generate speech audio. Before installing: - Understand what keys you provide: ELEVENLABS_API_KEY will cause text sent to the assistant to be sent to ElevenLabs for TTS (possible privacy and billing implications). Use a dedicated API key if you’re concerned about data exposure or costs. - Keep GATEWAY_URL pointed to a local gateway (default is ws://127.0.0.1:18789). Do not supply a token that grants broad access to a remote gateway unless you trust that endpoint; the client authenticates as a node with operator.write in the example handshake. - PORCUPINE_ACCESS_KEY is required for wake-word; Picovoice keys may expire. - The code creates temporary WAV files and asset files under the skill folder; ensure you run it from a directory you control. - The skill installs third-party Python packages and will (on first run) download Whisper models (~100–150MB or larger for bigger models). Consider installing and running in a virtualenv or isolated environment. - If you need stronger guarantees, review the included Python files yourself (we inspected them) or run the assistant in a sandboxed/isolated VM to monitor network calls and file writes. Overall: coherent and consistent with its stated purpose, but treat the API keys and gateway token as sensitive and verify the gateway target before use.

Type: OpenClaw Skill Name: openclaw-voice-assistant Version: 1.0.4 The skill bundle implements a voice assistant for OpenClaw, utilizing local speech-to-text (faster-whisper), wake word detection (Porcupine), and text-to-speech (ElevenLabs). All network communication is directed to the local OpenClaw gateway, ElevenLabs, and Picovoice, which are essential for its stated functionality. Sensitive API keys are loaded from `.env` and used appropriately by their respective client libraries. The `SKILL.md` provides user-facing instructions, including an optional manual step for auto-start, but contains no prompt injection attempts against the AI agent or evidence of stealthy persistence. All dependencies are legitimate, and no malicious execution, data exfiltration, or obfuscation was found.