Jarvis Wake-Word Assistant (OpenClaw)

A wake-word voice assistant for macOS:
“Jarvis” → record → transcribe → route to OpenClaw → speak the response.

---

✨ Features

• Wake-word detection via Picovoice Porcupine
• Local microphone recording
• Speech-to-text transcription using Whisper
• Command routing via OpenClaw CLI
• Text-to-speech using macOS say
• Environment-based secure configuration
• Optional Gemini fallback (via external keys)

---

📦 Requirements

• macOS (for built-in say)
• Python 3.10+
• A working microphone
• Picovoice Porcupine access key (required)

---

🛠 Installation

python3 -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install pvporcupine pyaudio openai-whisper

PyAudio on macOS

If installation fails:

brew install portaudio
pip install pyaudio

---

🔐 Configuration (Secrets via Environment Variables)

Required

export PICOVOICE_KEY="your_picovoice_key_here"

---

Optional Configuration

export WAKE_WORD="jarvis"
export WAKE_SENSITIVITY="0.6"
export WHISPER_MODEL="base"
export WHISPER_COMPUTE="auto"        # auto | cpu | cuda
export OPENCLAW_BIN="openclaw"
export TTS_VOICE=""                  # macOS voice name
export TTS_RATE="200"
export OLLAMA_HOST=""

---

🤖 Gemini Fallback (Optional)

Keys must NOT be committed.

Option 1 — Environment variable

export GEMINI_FALLBACK_KEYS="key1,key2,key3"

Option 2 — File (recommended)

export GEMINI_FALLBACK_KEYS_FILE="$HOME/.secrets/gemini_keys.txt"

One key per line:

# comment allowed
your_key_here
another_key_here

Optional:

export GEMINI_FALLBACK_MODEL="gemini-2.0-flash"

---

🚀 Running

python3 jarvis.py

Or:

chmod +x jarvis.py
./jarvis.py

---

🔒 Security Notes

Never commit:

• .env
• *.local
• *keys*.txt
• Any secret/token file
• ~/.openclaw/

Suggested .gitignore additions:

.env
*.local
*.secrets
*keys*.txt
**/.openclaw/*

All credentials must be injected at runtime via environment variables.

---

🧠 How It Works

1. Listens continuously for the configured wake word.
2. When detected, records audio until silence.
3. Transcribes audio using Whisper.
4. Sends the transcription to OpenClaw CLI.
5. Speaks the response using macOS TTS.

---

🧪 Troubleshooting

PICOVOICE_KEY is required

Make sure it is set:

echo $PICOVOICE_KEY

If empty, export it again.

---

PyAudio build errors

brew install portaudio
pip install --force-reinstall pyaudio